2014-02-26 19:11:54 +01:00
|
|
|
/*****************************************************************************
|
|
|
|
|
2016-06-21 14:21:03 +02:00
|
|
|
Copyright (c) 1995, 2016, Oracle and/or its affiliates. All Rights Reserved.
|
2021-01-07 11:18:13 +02:00
|
|
|
Copyright (c) 2013, 2021, MariaDB Corporation.
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
This program is free software; you can redistribute it and/or modify it under
|
|
|
|
the terms of the GNU General Public License as published by the Free Software
|
|
|
|
Foundation; version 2 of the License.
|
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful, but WITHOUT
|
|
|
|
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
|
|
|
|
FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License along with
|
|
|
|
this program; if not, write to the Free Software Foundation, Inc.,
|
2019-05-11 18:08:32 +03:00
|
|
|
51 Franklin Street, Fifth Floor, Boston, MA 02110-1335 USA
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
*****************************************************************************/
|
|
|
|
|
|
|
|
/**************************************************//**
|
|
|
|
@file include/buf0buf.h
|
|
|
|
The database buffer pool high-level routines
|
|
|
|
|
|
|
|
Created 11/5/1995 Heikki Tuuri
|
|
|
|
*******************************************************/
|
|
|
|
|
|
|
|
#ifndef buf0buf_h
|
|
|
|
#define buf0buf_h
|
|
|
|
|
2016-12-30 15:04:10 +02:00
|
|
|
/** Magic value to use instead of checksums when they are disabled */
|
|
|
|
#define BUF_NO_CHECKSUM_MAGIC 0xDEADBEEFUL
|
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
#include "fil0fil.h"
|
|
|
|
#include "mtr0types.h"
|
|
|
|
#include "buf0types.h"
|
2020-03-18 14:57:01 +03:00
|
|
|
#include "span.h"
|
2020-05-07 12:25:00 +03:00
|
|
|
#include "assume_aligned.h"
|
2016-08-12 11:17:45 +03:00
|
|
|
#ifndef UNIV_INNOCHECKSUM
|
2014-02-26 19:11:54 +01:00
|
|
|
#include "hash0hash.h"
|
|
|
|
#include "ut0byte.h"
|
|
|
|
#include "page0types.h"
|
|
|
|
#include "log0log.h"
|
2016-08-12 11:17:45 +03:00
|
|
|
#include "srv0srv.h"
|
|
|
|
#include <ostream>
|
|
|
|
|
|
|
|
// Forward declaration
|
|
|
|
struct fil_addr_t;
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
/** @name Modes for buf_page_get_gen */
|
|
|
|
/* @{ */
|
|
|
|
#define BUF_GET 10 /*!< get always */
|
|
|
|
#define BUF_GET_IF_IN_POOL 11 /*!< get if in pool */
|
|
|
|
#define BUF_PEEK_IF_IN_POOL 12 /*!< get if in pool, do not make
|
|
|
|
the block young in the LRU list */
|
|
|
|
#define BUF_GET_NO_LATCH 14 /*!< get and bufferfix, but
|
|
|
|
set no latch; we have
|
|
|
|
separated this case, because
|
|
|
|
it is error-prone programming
|
|
|
|
not to set a latch, and it
|
|
|
|
should be used with care */
|
|
|
|
#define BUF_GET_IF_IN_POOL_OR_WATCH 15
|
|
|
|
/*!< Get the page only if it's in the
|
|
|
|
buffer pool, if not then set a watch
|
|
|
|
on the page. */
|
|
|
|
#define BUF_GET_POSSIBLY_FREED 16
|
|
|
|
/*!< Like BUF_GET, but do not mind
|
|
|
|
if the file page has been freed. */
|
MDEV-13328 ALTER TABLE…DISCARD TABLESPACE takes a lot of time
With a big buffer pool that contains many data pages,
DISCARD TABLESPACE took a long time, because it would scan the
entire buffer pool to remove any pages that belong to the tablespace.
With a large buffer pool, this would take a lot of time, especially
when the table-to-discard is empty.
The minimum amount of work that DISCARD TABLESPACE must do is to
remove the pages of the to-be-discarded table from the
buf_pool->flush_list because any writes to the data file must be
prevented before the file is deleted.
If DISCARD TABLESPACE does not evict the pages from the buffer pool,
then IMPORT TABLESPACE must do it, because we must prevent pre-DISCARD,
not-yet-evicted pages from being mistaken for pages of the imported
tablespace.
It would not be a useful fix to simply move the buffer pool scan to
the IMPORT TABLESPACE step. What we can do is to actively evict those
pages that could be mistaken for imported pages. In this way, when
importing a small table into a big buffer pool, the import should
still run relatively fast.
Import is bypassing the buffer pool when reading pages for the
adjustment phase. In the adjustment phase, if a page exists in
the buffer pool, we could replace it with the page from the imported
file. Unfortunately I did not get this to work properly, so instead
we will simply evict any matching page from the buffer pool.
buf_page_get_gen(): Implement BUF_EVICT_IF_IN_POOL, a new mode
where the requested page will be evicted if it is found. There
must be no unwritten changes for the page.
buf_remove_t: Remove. Instead, use trx!=NULL to signify that a write
to file is desired, and use a separate parameter bool drop_ahi.
buf_LRU_flush_or_remove_pages(), fil_delete_tablespace():
Replace buf_remove_t.
buf_LRU_remove_pages(), buf_LRU_remove_all_pages(): Remove.
PageConverter::m_mtr: A dummy mini-transaction buffer
PageConverter::PageConverter(): Complete the member initialization list.
PageConverter::operator()(): Evict any 'shadow' pages from the
buffer pool so that pre-existing (garbage) pages cannot be mistaken
for pages that exist in the being-imported file.
row_discard_tablespace(): Remove a bogus comment that seems to
refer to IMPORT TABLESPACE, not DISCARD TABLESPACE.
2017-11-02 22:38:37 +02:00
|
|
|
#define BUF_EVICT_IF_IN_POOL 20 /*!< evict a clean block if found */
|
2014-02-26 19:11:54 +01:00
|
|
|
/* @} */
|
|
|
|
|
2016-12-30 15:04:10 +02:00
|
|
|
# ifdef UNIV_DEBUG
|
2016-08-12 11:17:45 +03:00
|
|
|
extern my_bool buf_disable_resize_buffer_pool_debug; /*!< if TRUE, resizing
|
|
|
|
buffer pool is not allowed. */
|
2016-12-30 15:04:10 +02:00
|
|
|
# endif /* UNIV_DEBUG */
|
2016-09-06 09:43:16 +03:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** buf_page_t::state() values, distinguishing buf_page_t and buf_block_t */
|
|
|
|
enum buf_page_state
|
|
|
|
{
|
|
|
|
/** available in buf_pool.free or buf_pool.watch */
|
|
|
|
BUF_BLOCK_NOT_USED,
|
|
|
|
/** allocated for something else than a file page */
|
|
|
|
BUF_BLOCK_MEMORY,
|
|
|
|
/** a previously allocated file page, in transit to NOT_USED */
|
|
|
|
BUF_BLOCK_REMOVE_HASH,
|
|
|
|
/** a buf_block_t that is also in buf_pool.LRU */
|
|
|
|
BUF_BLOCK_FILE_PAGE,
|
|
|
|
/** the buf_page_t of a ROW_FORMAT=COMPRESSED page
|
|
|
|
whose uncompressed page frame has been evicted */
|
|
|
|
BUF_BLOCK_ZIP_PAGE
|
2014-02-26 19:11:54 +01:00
|
|
|
};
|
|
|
|
|
|
|
|
/** This structure defines information we will fetch from each buffer pool. It
|
|
|
|
will be used to print table IO stats */
|
2020-02-12 14:45:21 +02:00
|
|
|
struct buf_pool_info_t
|
|
|
|
{
|
2014-02-26 19:11:54 +01:00
|
|
|
/* General buffer pool info */
|
|
|
|
ulint pool_size; /*!< Buffer Pool size in pages */
|
2020-03-18 21:48:00 +02:00
|
|
|
ulint lru_len; /*!< Length of buf_pool.LRU */
|
|
|
|
ulint old_lru_len; /*!< buf_pool.LRU_old_len */
|
|
|
|
ulint free_list_len; /*!< Length of buf_pool.free list */
|
|
|
|
ulint flush_list_len; /*!< Length of buf_pool.flush_list */
|
|
|
|
ulint n_pend_unzip; /*!< buf_pool.n_pend_unzip, pages
|
2014-02-26 19:11:54 +01:00
|
|
|
pending decompress */
|
2020-03-18 21:48:00 +02:00
|
|
|
ulint n_pend_reads; /*!< buf_pool.n_pend_reads, pages
|
2014-02-26 19:11:54 +01:00
|
|
|
pending read */
|
|
|
|
ulint n_pending_flush_lru; /*!< Pages pending flush in LRU */
|
|
|
|
ulint n_pending_flush_list; /*!< Pages pending flush in FLUSH
|
|
|
|
LIST */
|
|
|
|
ulint n_pages_made_young; /*!< number of pages made young */
|
|
|
|
ulint n_pages_not_made_young; /*!< number of pages not made young */
|
2020-03-18 21:48:00 +02:00
|
|
|
ulint n_pages_read; /*!< buf_pool.n_pages_read */
|
|
|
|
ulint n_pages_created; /*!< buf_pool.n_pages_created */
|
|
|
|
ulint n_pages_written; /*!< buf_pool.n_pages_written */
|
|
|
|
ulint n_page_gets; /*!< buf_pool.n_page_gets */
|
|
|
|
ulint n_ra_pages_read_rnd; /*!< buf_pool.n_ra_pages_read_rnd,
|
2014-02-26 19:11:54 +01:00
|
|
|
number of pages readahead */
|
2020-03-18 21:48:00 +02:00
|
|
|
ulint n_ra_pages_read; /*!< buf_pool.n_ra_pages_read, number
|
2014-02-26 19:11:54 +01:00
|
|
|
of pages readahead */
|
2020-03-18 21:48:00 +02:00
|
|
|
ulint n_ra_pages_evicted; /*!< buf_pool.n_ra_pages_evicted,
|
2014-02-26 19:11:54 +01:00
|
|
|
number of readahead pages evicted
|
|
|
|
without access */
|
|
|
|
ulint n_page_get_delta; /*!< num of buffer pool page gets since
|
|
|
|
last printout */
|
|
|
|
|
|
|
|
/* Buffer pool access stats */
|
|
|
|
double page_made_young_rate; /*!< page made young rate in pages
|
|
|
|
per second */
|
|
|
|
double page_not_made_young_rate;/*!< page not made young rate
|
|
|
|
in pages per second */
|
|
|
|
double pages_read_rate; /*!< num of pages read per second */
|
|
|
|
double pages_created_rate; /*!< num of pages create per second */
|
|
|
|
double pages_written_rate; /*!< num of pages written per second */
|
|
|
|
ulint page_read_delta; /*!< num of pages read since last
|
|
|
|
printout */
|
|
|
|
ulint young_making_delta; /*!< num of pages made young since
|
|
|
|
last printout */
|
|
|
|
ulint not_young_making_delta; /*!< num of pages not make young since
|
|
|
|
last printout */
|
|
|
|
|
|
|
|
/* Statistics about read ahead algorithm. */
|
|
|
|
double pages_readahead_rnd_rate;/*!< random readahead rate in pages per
|
|
|
|
second */
|
|
|
|
double pages_readahead_rate; /*!< readahead rate in pages per
|
|
|
|
second */
|
|
|
|
double pages_evicted_rate; /*!< rate of readahead page evicted
|
|
|
|
without access, in pages per second */
|
|
|
|
|
|
|
|
/* Stats about LRU eviction */
|
2020-03-18 21:48:00 +02:00
|
|
|
ulint unzip_lru_len; /*!< length of buf_pool.unzip_LRU
|
2014-02-26 19:11:54 +01:00
|
|
|
list */
|
|
|
|
/* Counters for LRU policy */
|
|
|
|
ulint io_sum; /*!< buf_LRU_stat_sum.io */
|
|
|
|
ulint io_cur; /*!< buf_LRU_stat_cur.io, num of IO
|
|
|
|
for current interval */
|
|
|
|
ulint unzip_sum; /*!< buf_LRU_stat_sum.unzip */
|
|
|
|
ulint unzip_cur; /*!< buf_LRU_stat_cur.unzip, num
|
|
|
|
pages decompressed in current
|
|
|
|
interval */
|
|
|
|
};
|
2017-08-04 13:57:26 +03:00
|
|
|
#endif /* !UNIV_INNOCHECKSUM */
|
2014-02-26 19:11:54 +01:00
|
|
|
|
2016-08-12 11:17:45 +03:00
|
|
|
/** Print the given page_id_t object.
|
|
|
|
@param[in,out] out the output stream
|
|
|
|
@param[in] page_id the page_id_t object to be printed
|
|
|
|
@return the output stream */
|
|
|
|
std::ostream&
|
|
|
|
operator<<(
|
|
|
|
std::ostream& out,
|
2018-10-18 18:23:12 +03:00
|
|
|
const page_id_t page_id);
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2017-08-04 13:57:26 +03:00
|
|
|
#ifndef UNIV_INNOCHECKSUM
|
2014-02-26 19:11:54 +01:00
|
|
|
/*********************************************************************//**
|
|
|
|
Gets the current size of buffer buf_pool in bytes.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return size in bytes */
|
2014-02-26 19:11:54 +01:00
|
|
|
UNIV_INLINE
|
|
|
|
ulint
|
|
|
|
buf_pool_get_curr_size(void);
|
|
|
|
/*========================*/
|
|
|
|
|
|
|
|
/********************************************************************//**
|
|
|
|
Allocates a buf_page_t descriptor. This function must succeed. In case
|
|
|
|
of failure we assert in this function. */
|
|
|
|
UNIV_INLINE
|
|
|
|
buf_page_t*
|
|
|
|
buf_page_alloc_descriptor(void)
|
|
|
|
/*===========================*/
|
2016-06-21 14:21:03 +02:00
|
|
|
MY_ATTRIBUTE((malloc));
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
Free a buf_page_t descriptor. */
|
|
|
|
UNIV_INLINE
|
|
|
|
void
|
|
|
|
buf_page_free_descriptor(
|
|
|
|
/*=====================*/
|
|
|
|
buf_page_t* bpage) /*!< in: bpage descriptor to free. */
|
2016-06-21 14:21:03 +02:00
|
|
|
MY_ATTRIBUTE((nonnull));
|
2014-02-26 19:11:54 +01:00
|
|
|
|
2020-02-12 14:45:21 +02:00
|
|
|
/** Allocate a buffer block.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return own: the allocated block, in state BUF_BLOCK_MEMORY */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline buf_block_t *buf_block_alloc();
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
Frees a buffer block which does not contain a file page. */
|
|
|
|
UNIV_INLINE
|
|
|
|
void
|
|
|
|
buf_block_free(
|
|
|
|
/*===========*/
|
|
|
|
buf_block_t* block); /*!< in, own: block to be freed */
|
2016-12-30 15:04:10 +02:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/**************************************************************//**
|
|
|
|
NOTE! The following macros should be used instead of buf_page_get_gen,
|
|
|
|
to improve debugging. Only values RW_S_LATCH and RW_X_LATCH are allowed
|
|
|
|
in LA! */
|
2016-08-12 11:17:45 +03:00
|
|
|
#define buf_page_get(ID, SIZE, LA, MTR) \
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
buf_page_get_gen(ID, SIZE, LA, NULL, BUF_GET, __FILE__, __LINE__, MTR)
|
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/**************************************************************//**
|
|
|
|
Use these macros to bufferfix a page with no latching. Remember not to
|
|
|
|
read the contents of the page unless you know it is safe. Do not modify
|
|
|
|
the contents of the page! We have separated this case, because it is
|
|
|
|
error-prone programming not to set a latch, and it should be used
|
|
|
|
with care. */
|
2016-08-12 11:17:45 +03:00
|
|
|
#define buf_page_get_with_no_latch(ID, SIZE, MTR) \
|
|
|
|
buf_page_get_gen(ID, SIZE, RW_NO_LATCH, NULL, BUF_GET_NO_LATCH, \
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
__FILE__, __LINE__, MTR)
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
This is the general function used to get optimistic access to a database
|
|
|
|
page.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return TRUE if success */
|
2014-02-26 19:11:54 +01:00
|
|
|
ibool
|
|
|
|
buf_page_optimistic_get(
|
|
|
|
/*====================*/
|
|
|
|
ulint rw_latch,/*!< in: RW_S_LATCH, RW_X_LATCH */
|
|
|
|
buf_block_t* block, /*!< in: guessed block */
|
|
|
|
ib_uint64_t modify_clock,/*!< in: modify clock value */
|
|
|
|
const char* file, /*!< in: file name */
|
2017-03-01 08:27:39 +02:00
|
|
|
unsigned line, /*!< in: line where called */
|
2014-02-26 19:11:54 +01:00
|
|
|
mtr_t* mtr); /*!< in: mini-transaction */
|
|
|
|
|
2016-08-12 11:17:45 +03:00
|
|
|
/** Given a tablespace id and page number tries to get that page. If the
|
2014-02-26 19:11:54 +01:00
|
|
|
page is not in the buffer pool it is not loaded and NULL is returned.
|
2016-08-12 11:17:45 +03:00
|
|
|
Suitable for using when holding the lock_sys_t::mutex.
|
|
|
|
@param[in] page_id page id
|
|
|
|
@param[in] file file name
|
|
|
|
@param[in] line line where called
|
|
|
|
@param[in] mtr mini-transaction
|
|
|
|
@return pointer to a page or NULL */
|
2014-12-22 16:53:17 +02:00
|
|
|
buf_block_t*
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_page_try_get_func(
|
2018-10-18 18:23:12 +03:00
|
|
|
const page_id_t page_id,
|
2016-08-12 11:17:45 +03:00
|
|
|
const char* file,
|
2017-03-01 08:27:39 +02:00
|
|
|
unsigned line,
|
2016-08-12 11:17:45 +03:00
|
|
|
mtr_t* mtr);
|
|
|
|
|
|
|
|
/** Tries to get a page.
|
|
|
|
If the page is not in the buffer pool it is not loaded. Suitable for using
|
|
|
|
when holding the lock_sys_t::mutex.
|
|
|
|
@param[in] page_id page identifier
|
|
|
|
@param[in] mtr mini-transaction
|
|
|
|
@return the page if in buffer pool, NULL if not */
|
|
|
|
#define buf_page_try_get(page_id, mtr) \
|
|
|
|
buf_page_try_get_func((page_id), __FILE__, __LINE__, mtr);
|
|
|
|
|
|
|
|
/** Get read access to a compressed page (usually of type
|
2014-02-26 19:11:54 +01:00
|
|
|
FIL_PAGE_TYPE_ZBLOB or FIL_PAGE_TYPE_ZBLOB2).
|
|
|
|
The page must be released with buf_page_release_zip().
|
|
|
|
NOTE: the page is not protected by any latch. Mutual exclusion has to
|
|
|
|
be implemented at a higher level. In other words, all possible
|
|
|
|
accesses to a given page through this function must be protected by
|
|
|
|
the same set of mutexes or latches.
|
2016-08-12 11:17:45 +03:00
|
|
|
@param[in] page_id page id
|
2019-02-06 19:50:11 +02:00
|
|
|
@param[in] zip_size ROW_FORMAT=COMPRESSED page size
|
2016-08-12 11:17:45 +03:00
|
|
|
@return pointer to the block */
|
2019-02-06 19:50:11 +02:00
|
|
|
buf_page_t* buf_page_get_zip(const page_id_t page_id, ulint zip_size);
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2020-03-30 18:52:17 +03:00
|
|
|
/** Get access to a database page. Buffered redo log may be applied.
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
@param[in] page_id page id
|
|
|
|
@param[in] zip_size ROW_FORMAT=COMPRESSED page size, or 0
|
|
|
|
@param[in] rw_latch RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH
|
|
|
|
@param[in] guess guessed block or NULL
|
|
|
|
@param[in] mode BUF_GET, BUF_GET_IF_IN_POOL,
|
2016-08-12 11:17:45 +03:00
|
|
|
BUF_PEEK_IF_IN_POOL, BUF_GET_NO_LATCH, or BUF_GET_IF_IN_POOL_OR_WATCH
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
@param[in] file file name
|
|
|
|
@param[in] line line where called
|
|
|
|
@param[in] mtr mini-transaction
|
|
|
|
@param[out] err DB_SUCCESS or error code
|
|
|
|
@param[in] allow_ibuf_merge Allow change buffer merge while
|
|
|
|
reading the pages from file.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return pointer to the block or NULL */
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_block_t*
|
|
|
|
buf_page_get_gen(
|
2018-10-18 18:23:12 +03:00
|
|
|
const page_id_t page_id,
|
2019-02-06 19:50:11 +02:00
|
|
|
ulint zip_size,
|
2016-08-12 11:17:45 +03:00
|
|
|
ulint rw_latch,
|
|
|
|
buf_block_t* guess,
|
|
|
|
ulint mode,
|
|
|
|
const char* file,
|
2017-03-01 08:27:39 +02:00
|
|
|
unsigned line,
|
2016-08-12 11:17:45 +03:00
|
|
|
mtr_t* mtr,
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
dberr_t* err = NULL,
|
|
|
|
bool allow_ibuf_merge = false);
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2020-03-30 18:52:17 +03:00
|
|
|
/** This is the low level function used to get access to a database page.
|
|
|
|
@param[in] page_id page id
|
|
|
|
@param[in] zip_size ROW_FORMAT=COMPRESSED page size, or 0
|
|
|
|
@param[in] rw_latch RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH
|
|
|
|
@param[in] guess guessed block or NULL
|
|
|
|
@param[in] mode BUF_GET, BUF_GET_IF_IN_POOL,
|
2020-03-23 16:37:44 +05:30
|
|
|
BUF_PEEK_IF_IN_POOL, BUF_GET_NO_LATCH, or BUF_GET_IF_IN_POOL_OR_WATCH
|
2020-03-30 18:52:17 +03:00
|
|
|
@param[in] file file name
|
|
|
|
@param[in] line line where called
|
|
|
|
@param[in] mtr mini-transaction
|
|
|
|
@param[out] err DB_SUCCESS or error code
|
|
|
|
@param[in] allow_ibuf_merge Allow change buffer merge to happen
|
|
|
|
while reading the page from file
|
|
|
|
then it makes sure that it does merging of change buffer changes while
|
|
|
|
reading the page from file.
|
2020-03-23 16:37:44 +05:30
|
|
|
@return pointer to the block or NULL */
|
|
|
|
buf_block_t*
|
|
|
|
buf_page_get_low(
|
|
|
|
const page_id_t page_id,
|
2020-03-30 14:50:23 +03:00
|
|
|
ulint zip_size,
|
2020-03-23 16:37:44 +05:30
|
|
|
ulint rw_latch,
|
|
|
|
buf_block_t* guess,
|
|
|
|
ulint mode,
|
|
|
|
const char* file,
|
|
|
|
unsigned line,
|
|
|
|
mtr_t* mtr,
|
2020-03-30 18:52:17 +03:00
|
|
|
dberr_t* err,
|
|
|
|
bool allow_ibuf_merge);
|
2020-03-23 16:37:44 +05:30
|
|
|
|
2019-02-06 19:50:11 +02:00
|
|
|
/** Initialize a page in the buffer pool. The page is usually not read
|
2014-02-26 19:11:54 +01:00
|
|
|
from a file even if it cannot be found in the buffer buf_pool. This is one
|
|
|
|
of the functions which perform to a block a state transition NOT_USED =>
|
|
|
|
FILE_PAGE (the other is buf_page_get_gen).
|
2020-06-11 22:52:47 +05:30
|
|
|
@param[in,out] space space object
|
|
|
|
@param[in] offset offset of the tablespace
|
2019-02-06 19:50:11 +02:00
|
|
|
@param[in] zip_size ROW_FORMAT=COMPRESSED page size, or 0
|
|
|
|
@param[in,out] mtr mini-transaction
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
@param[in,out] free_block pre-allocated buffer block
|
2016-08-12 11:17:45 +03:00
|
|
|
@return pointer to the block, page bufferfixed */
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_block_t*
|
2020-06-11 22:52:47 +05:30
|
|
|
buf_page_create(fil_space_t *space, uint32_t offset,
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
ulint zip_size, mtr_t *mtr, buf_block_t *free_block);
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
Releases a compressed-only page acquired with buf_page_get_zip(). */
|
|
|
|
UNIV_INLINE
|
|
|
|
void
|
|
|
|
buf_page_release_zip(
|
|
|
|
/*=================*/
|
|
|
|
buf_page_t* bpage); /*!< in: buffer block */
|
|
|
|
/********************************************************************//**
|
2016-08-12 11:17:45 +03:00
|
|
|
Releases a latch, if specified. */
|
2014-02-26 19:11:54 +01:00
|
|
|
UNIV_INLINE
|
|
|
|
void
|
2016-08-12 11:17:45 +03:00
|
|
|
buf_page_release_latch(
|
|
|
|
/*=====================*/
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_block_t* block, /*!< in: buffer block */
|
|
|
|
ulint rw_latch); /*!< in: RW_S_LATCH, RW_X_LATCH,
|
|
|
|
RW_NO_LATCH */
|
2020-06-08 13:14:27 +03:00
|
|
|
/** Move a block to the start of the LRU list. */
|
|
|
|
void buf_page_make_young(buf_page_t *bpage);
|
MDEV-15528 Punch holes when pages are freed
When a InnoDB data file page is freed, its contents becomes garbage,
and any storage allocated in the data file is wasted. During flushing,
InnoDB initializes the page with zeros if scrubbing is enabled. If the
tablespace is compressed then InnoDB should punch a hole else ignore the
flushing of the freed page.
buf_page_t:
- Replaced the variable file_page_was_freed, init_on_flush in buf_page_t
with status enum variable.
- Changed all debug assert of file_page_was_freed to DBUG_ASSERT
of buf_page_t::status
Removed buf_page_set_file_page_was_freed(),
buf_page_reset_file_page_was_freed().
buf_page_free(): Newly added function which takes X-lock on the page
before marking the status as FREED. So that InnoDB flush handler can
avoid concurrent flush of the freed page. Also while flushing the page,
InnoDB make sure that redo log which does freeing of the page also written
to the disk. Currently, this function only marks the page as FREED if
it is in buffer pool
buf_flush_freed_page(): Newly added function which initializes zeros
asynchorously if innodb_immediate_scrub_data_uncompressed is enabled.
Punch a hole to the file synchorously if page_compressed is enabled.
Reset the io_fix to NORMAL. Release the block from flush list and
associated mutex before writing zeros or punch a hole to the file.
buf_flush_page(): Removed the unnecessary usage of temporary
variable "flush"
fil_io(): Introduce new parameter called punch_hole. It allows fil_io()
to punch the hole to the file for the given offset.
buf_page_create(): Let the callers assign buf_page_t::status.
Every caller should eventually invoke mtr_t::init().
fsp_page_create(): Remove the unused mtr_t parameter.
In all other callers of buf_page_create() except fsp_page_create(),
before invoking mtr_t::init(), invoke
mtr_t::sx_latch_at_savepoint() or mtr_t::x_latch_at_savepoint().
mtr_t::init(): Initialize buf_page_t::status also for the temporary
tablespace (when redo logging is disabled), to avoid assertion failures.
2020-03-09 13:25:33 +05:30
|
|
|
/** Mark the page status as FREED for the given tablespace id and
|
|
|
|
page number. If the page is not in buffer pool then ignore it.
|
2020-12-18 17:12:57 +02:00
|
|
|
@param[in,out] space tablespace
|
|
|
|
@param[in] page page number
|
MDEV-15528 Punch holes when pages are freed
When a InnoDB data file page is freed, its contents becomes garbage,
and any storage allocated in the data file is wasted. During flushing,
InnoDB initializes the page with zeros if scrubbing is enabled. If the
tablespace is compressed then InnoDB should punch a hole else ignore the
flushing of the freed page.
buf_page_t:
- Replaced the variable file_page_was_freed, init_on_flush in buf_page_t
with status enum variable.
- Changed all debug assert of file_page_was_freed to DBUG_ASSERT
of buf_page_t::status
Removed buf_page_set_file_page_was_freed(),
buf_page_reset_file_page_was_freed().
buf_page_free(): Newly added function which takes X-lock on the page
before marking the status as FREED. So that InnoDB flush handler can
avoid concurrent flush of the freed page. Also while flushing the page,
InnoDB make sure that redo log which does freeing of the page also written
to the disk. Currently, this function only marks the page as FREED if
it is in buffer pool
buf_flush_freed_page(): Newly added function which initializes zeros
asynchorously if innodb_immediate_scrub_data_uncompressed is enabled.
Punch a hole to the file synchorously if page_compressed is enabled.
Reset the io_fix to NORMAL. Release the block from flush list and
associated mutex before writing zeros or punch a hole to the file.
buf_flush_page(): Removed the unnecessary usage of temporary
variable "flush"
fil_io(): Introduce new parameter called punch_hole. It allows fil_io()
to punch the hole to the file for the given offset.
buf_page_create(): Let the callers assign buf_page_t::status.
Every caller should eventually invoke mtr_t::init().
fsp_page_create(): Remove the unused mtr_t parameter.
In all other callers of buf_page_create() except fsp_page_create(),
before invoking mtr_t::init(), invoke
mtr_t::sx_latch_at_savepoint() or mtr_t::x_latch_at_savepoint().
mtr_t::init(): Initialize buf_page_t::status also for the temporary
tablespace (when redo logging is disabled), to avoid assertion failures.
2020-03-09 13:25:33 +05:30
|
|
|
@param[in,out] mtr mini-transaction
|
|
|
|
@param[in] file file name
|
|
|
|
@param[in] line line where called */
|
2020-12-18 17:12:57 +02:00
|
|
|
void buf_page_free(fil_space_t *space, uint32_t page, mtr_t *mtr,
|
|
|
|
const char *file, unsigned line);
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
Reads the freed_page_clock of a buffer block.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return freed_page_clock */
|
2014-02-26 19:11:54 +01:00
|
|
|
UNIV_INLINE
|
2017-03-01 08:27:39 +02:00
|
|
|
unsigned
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_page_get_freed_page_clock(
|
|
|
|
/*==========================*/
|
|
|
|
const buf_page_t* bpage) /*!< in: block */
|
2016-09-06 09:43:16 +03:00
|
|
|
MY_ATTRIBUTE((warn_unused_result));
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
Reads the freed_page_clock of a buffer block.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return freed_page_clock */
|
2014-02-26 19:11:54 +01:00
|
|
|
UNIV_INLINE
|
2017-03-01 08:27:39 +02:00
|
|
|
unsigned
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_block_get_freed_page_clock(
|
|
|
|
/*===========================*/
|
|
|
|
const buf_block_t* block) /*!< in: block */
|
2016-09-06 09:43:16 +03:00
|
|
|
MY_ATTRIBUTE((warn_unused_result));
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
/** Determine if a block is still close enough to the MRU end of the LRU list
|
2014-02-26 19:11:54 +01:00
|
|
|
meaning that it is not in danger of getting evicted and also implying
|
|
|
|
that it has been accessed recently.
|
|
|
|
Note that this is for heuristics only and does not reserve buffer pool
|
|
|
|
mutex.
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
@param[in] bpage buffer pool page
|
|
|
|
@return whether bpage is close to MRU end of LRU */
|
2020-02-12 14:45:21 +02:00
|
|
|
inline bool buf_page_peek_if_young(const buf_page_t *bpage);
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
|
|
|
|
/** Determine if a block should be moved to the start of the LRU list if
|
|
|
|
there is danger of dropping from the buffer pool.
|
|
|
|
@param[in] bpage buffer pool page
|
|
|
|
@return true if bpage should be made younger */
|
2020-02-12 14:45:21 +02:00
|
|
|
inline bool buf_page_peek_if_too_old(const buf_page_t *bpage);
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
|
|
|
|
/** Move a page to the start of the buffer pool LRU list if it is too old.
|
|
|
|
@param[in,out] bpage buffer pool page */
|
2020-02-12 14:45:21 +02:00
|
|
|
inline void buf_page_make_young_if_needed(buf_page_t *bpage)
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
{
|
2020-02-12 14:45:21 +02:00
|
|
|
if (UNIV_UNLIKELY(buf_page_peek_if_too_old(bpage))) {
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
buf_page_make_young(bpage);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
Increments the modify clock of a frame by 1. The caller must (1) own the
|
2020-03-18 21:48:00 +02:00
|
|
|
buf_pool.mutex and block bufferfix count has to be zero, (2) or own an x-lock
|
2014-02-26 19:11:54 +01:00
|
|
|
on the block. */
|
|
|
|
UNIV_INLINE
|
|
|
|
void
|
|
|
|
buf_block_modify_clock_inc(
|
|
|
|
/*=======================*/
|
|
|
|
buf_block_t* block); /*!< in: block */
|
|
|
|
/********************************************************************//**
|
|
|
|
Returns the value of the modify clock. The caller must have an s-lock
|
|
|
|
or x-lock on the block.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return value */
|
2014-02-26 19:11:54 +01:00
|
|
|
UNIV_INLINE
|
|
|
|
ib_uint64_t
|
|
|
|
buf_block_get_modify_clock(
|
|
|
|
/*=======================*/
|
|
|
|
buf_block_t* block); /*!< in: block */
|
|
|
|
/*******************************************************************//**
|
|
|
|
Increments the bufferfix count. */
|
|
|
|
UNIV_INLINE
|
|
|
|
void
|
|
|
|
buf_block_buf_fix_inc_func(
|
|
|
|
/*=======================*/
|
2016-08-12 11:17:45 +03:00
|
|
|
# ifdef UNIV_DEBUG
|
2014-02-26 19:11:54 +01:00
|
|
|
const char* file, /*!< in: file name */
|
2017-03-01 08:27:39 +02:00
|
|
|
unsigned line, /*!< in: line */
|
2016-08-12 11:17:45 +03:00
|
|
|
# endif /* UNIV_DEBUG */
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_block_t* block) /*!< in/out: block to bufferfix */
|
2016-06-21 14:21:03 +02:00
|
|
|
MY_ATTRIBUTE((nonnull));
|
2014-05-05 18:20:28 +02:00
|
|
|
|
2016-08-12 11:17:45 +03:00
|
|
|
# ifdef UNIV_DEBUG
|
2014-02-26 19:11:54 +01:00
|
|
|
/** Increments the bufferfix count.
|
2016-08-12 11:17:45 +03:00
|
|
|
@param[in,out] b block to bufferfix
|
|
|
|
@param[in] f file name where requested
|
|
|
|
@param[in] l line number where requested */
|
2016-12-30 15:04:10 +02:00
|
|
|
# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(f,l,b)
|
2016-08-12 11:17:45 +03:00
|
|
|
# else /* UNIV_DEBUG */
|
2014-02-26 19:11:54 +01:00
|
|
|
/** Increments the bufferfix count.
|
2016-08-12 11:17:45 +03:00
|
|
|
@param[in,out] b block to bufferfix
|
|
|
|
@param[in] f file name where requested
|
|
|
|
@param[in] l line number where requested */
|
2016-12-30 15:04:10 +02:00
|
|
|
# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(b)
|
2016-08-12 11:17:45 +03:00
|
|
|
# endif /* UNIV_DEBUG */
|
|
|
|
#endif /* !UNIV_INNOCHECKSUM */
|
|
|
|
|
2020-03-18 14:57:01 +03:00
|
|
|
/** Check if a buffer is all zeroes.
|
|
|
|
@param[in] buf data to check
|
|
|
|
@return whether the buffer is all zeroes */
|
|
|
|
bool buf_is_zeroes(st_::span<const byte> buf);
|
MDEV-18644: Support full_crc32 for page_compressed
This is a follow-up task to MDEV-12026, which introduced
innodb_checksum_algorithm=full_crc32 and a simpler page format.
MDEV-12026 did not enable full_crc32 for page_compressed tables,
which we will be doing now.
This is joint work with Thirunarayanan Balathandayuthapani.
For innodb_checksum_algorithm=full_crc32 we change the
page_compressed format as follows:
FIL_PAGE_TYPE: The most significant bit will be set to indicate
page_compressed format. The least significant bits will contain
the compressed page size, rounded up to a multiple of 256 bytes.
The checksum will be stored in the last 4 bytes of the page
(whether it is the full page or a page_compressed page whose
size is determined by FIL_PAGE_TYPE), covering all preceding
bytes of the page. If encryption is used, then the page will
be encrypted between compression and computing the checksum.
For page_compressed, FIL_PAGE_LSN will not be repeated at
the end of the page.
FSP_SPACE_FLAGS (already implemented as part of MDEV-12026):
We will store the innodb_compression_algorithm that may be used
to compress pages. Previously, the choice of algorithm was written
to each compressed data page separately, and one would be unable
to know in advance which compression algorithm(s) are used.
fil_space_t::full_crc32_page_compressed_len(): Determine if the
page_compressed algorithm of the tablespace needs to know the
exact length of the compressed data. If yes, we will reserve and
write an extra byte for this right before the checksum.
buf_page_is_compressed(): Determine if a page uses page_compressed
(in any innodb_checksum_algorithm).
fil_page_decompress(): Pass also fil_space_t::flags so that the
format can be determined.
buf_page_is_zeroes(): Check if a page is full of zero bytes.
buf_page_full_crc32_is_corrupted(): Renamed from
buf_encrypted_full_crc32_page_is_corrupted(). For full_crc32,
we always simply validate the checksum to the page contents,
while the physical page size is explicitly specified by an
unencrypted part of the page header.
buf_page_full_crc32_size(): Determine the size of a full_crc32 page.
buf_dblwr_check_page_lsn(): Make this a debug-only function, because
it involves potentially costly lookups of fil_space_t.
create_table_info_t::check_table_options(),
ha_innobase::check_if_supported_inplace_alter(): Do allow the creation
of SPATIAL INDEX with full_crc32 also when page_compressed is used.
commit_cache_norebuild(): Preserve the compression algorithm when
updating the page_compression_level.
dict_tf_to_fsp_flags(): Set the flags for page compression algorithm.
FIXME: Maybe there should be a table option page_compression_algorithm
and a session variable to back it?
2019-03-18 14:08:43 +02:00
|
|
|
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
/** Checks if the page is in crc32 checksum format.
|
2017-03-30 12:48:42 +02:00
|
|
|
@param[in] read_buf database page
|
|
|
|
@param[in] checksum_field1 new checksum field
|
|
|
|
@param[in] checksum_field2 old checksum field
|
|
|
|
@return true if the page is in crc32 checksum format. */
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
bool
|
|
|
|
buf_page_is_checksum_valid_crc32(
|
2017-03-30 12:48:42 +02:00
|
|
|
const byte* read_buf,
|
|
|
|
ulint checksum_field1,
|
MDEV-17958 Make bug-endian innodb_checksum_algorithm=crc32 optional
In MySQL 5.7, it was noticed that files are not portable between
big-endian and little-endian processor architectures
(such as SPARC and x86), because the original implementation of
innodb_checksum_algorithm=crc32 was not byte order agnostic.
A byte order agnostic implementation of innodb_checksum_algorithm=crc32
was only added to MySQL 5.7, not backported to 5.6. Consequently,
MariaDB Server versions 10.0 and 10.1 only contain the CRC-32C
implementation that works incorrectly on big-endian architectures,
and MariaDB Server 10.2.2 got the byte-order agnostic CRC-32C
implementation from MySQL 5.7.
MySQL 5.7 introduced a "legacy crc32" variant that is functionally
equivalent to the big-endian version of the original crc32 implementation.
Thanks to this variant, old data files can be transferred from big-endian
systems to newer versions.
Introducing new variants of checksum algorithms (without introducing
new names for them, or something on the pages themselves to identify
the algorithm) generally is a bad idea, because each checksum algorithm
is like a lottery ticket. The more algorithms you try, the more likely
it will be for the checksum to match on a corrupted page.
So, essentially MySQL 5.7 weakened innodb_checksum_algorithm=crc32,
and MariaDB 10.2.2 inherited this weakening.
We introduce a build option that together with MDEV-17957
makes innodb_checksum_algorithm=strict_crc32 strict again
by only allowing one variant of the checksum to match.
WITH_INNODB_BUG_ENDIAN_CRC32: A new cmake option for enabling the
bug-compatible "legacy crc32" checksum. This is only enabled on
big-endian systems by default, to facilitate an upgrade from
MariaDB 10.0 or 10.1. Checked by #ifdef INNODB_BUG_ENDIAN_CRC32.
ut_crc32_byte_by_byte: Remove (unused function).
legacy_big_endian_checksum: Remove. This variable seems to have
unnecessarily complicated the logic. When the weakening is enabled,
we must always fall back to the buggy checksum.
buf_page_check_crc32(): A helper function to compute one or
two CRC-32C variants.
2018-12-13 17:57:10 +02:00
|
|
|
ulint checksum_field2)
|
2017-03-30 12:48:42 +02:00
|
|
|
MY_ATTRIBUTE((nonnull(1), warn_unused_result));
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
|
|
|
|
/** Checks if the page is in innodb checksum format.
|
|
|
|
@param[in] read_buf database page
|
|
|
|
@param[in] checksum_field1 new checksum field
|
|
|
|
@param[in] checksum_field2 old checksum field
|
2017-03-30 12:48:42 +02:00
|
|
|
@return true if the page is in innodb checksum format. */
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
bool
|
|
|
|
buf_page_is_checksum_valid_innodb(
|
2017-03-30 12:48:42 +02:00
|
|
|
const byte* read_buf,
|
|
|
|
ulint checksum_field1,
|
2017-04-07 15:01:27 +02:00
|
|
|
ulint checksum_field2)
|
2017-03-30 12:48:42 +02:00
|
|
|
MY_ATTRIBUTE((nonnull(1), warn_unused_result));
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
|
|
|
|
/** Checks if the page is in none checksum format.
|
|
|
|
@param[in] read_buf database page
|
|
|
|
@param[in] checksum_field1 new checksum field
|
|
|
|
@param[in] checksum_field2 old checksum field
|
2017-03-30 12:48:42 +02:00
|
|
|
@return true if the page is in none checksum format. */
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
bool
|
|
|
|
buf_page_is_checksum_valid_none(
|
2017-03-30 12:48:42 +02:00
|
|
|
const byte* read_buf,
|
|
|
|
ulint checksum_field1,
|
2017-04-07 15:01:27 +02:00
|
|
|
ulint checksum_field2)
|
2017-03-30 12:48:42 +02:00
|
|
|
MY_ATTRIBUTE((nonnull(1), warn_unused_result));
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
|
MDEV-12602 InnoDB: Failing assertion: space->n_pending_ops == 0
This fixes a regression caused by MDEV-12428.
When we introduced a variant of fil_space_acquire() that could
increment space->n_pending_ops after space->stop_new_ops was set,
the logic of fil_check_pending_operations() was broken.
fil_space_t::n_pending_ios: A new field to track read or write
access from the buffer pool routines immediately before a block
write or after a block read in the file system.
fil_space_acquire_for_io(), fil_space_release_for_io(): Similar
to fil_space_acquire_silent() and fil_space_release(), but
modify fil_space_t::n_pending_ios instead of fil_space_t::n_pending_ops.
Adjust a number of places accordingly, and remove some redundant
tablespace lookups.
The following parts of this fix differ from the 10.2 version of this fix:
buf_page_get_corrupt(): Add a tablespace parameter.
In 10.2, we already had a two-phase process of freeing fil_space objects
(first, fil_space_detach(), then release fil_system->mutex, and finally
free the fil_space and fil_node objects).
fil_space_free_and_mutex_exit(): Renamed from fil_space_free().
Detach the tablespace from the fil_system cache, release the
fil_system->mutex, and then wait for space->n_pending_ios to reach 0,
to avoid accessing freed data in a concurrent thread.
During the wait, future calls to fil_space_acquire_for_io() will
not find this tablespace, and the count can only be decremented to 0,
at which point it is safe to free the objects.
fil_node_free_part1(), fil_node_free_part2(): Refactored from
fil_node_free().
2017-04-28 03:20:49 +03:00
|
|
|
/** Check if a page is corrupt.
|
2017-05-06 14:36:46 +03:00
|
|
|
@param[in] check_lsn whether the LSN should be checked
|
2016-08-12 11:17:45 +03:00
|
|
|
@param[in] read_buf database page
|
MDEV-12026: Implement innodb_checksum_algorithm=full_crc32
MariaDB data-at-rest encryption (innodb_encrypt_tables)
had repurposed the same unused data field that was repurposed
in MySQL 5.7 (and MariaDB 10.2) for the Split Sequence Number (SSN)
field of SPATIAL INDEX. Because of this, MariaDB was unable to
support encryption on SPATIAL INDEX pages.
Furthermore, InnoDB page checksums skipped some bytes, and there
are multiple variations and checksum algorithms. By default,
InnoDB accepts all variations of all algorithms that ever existed.
This unnecessarily weakens the page checksums.
We hereby introduce two more innodb_checksum_algorithm variants
(full_crc32, strict_full_crc32) that are special in a way:
When either setting is active, newly created data files will
carry a flag (fil_space_t::full_crc32()) that indicates that
all pages of the file will use a full CRC-32C checksum over the
entire page contents (excluding the bytes where the checksum
is stored, at the very end of the page). Such files will always
use that checksum, no matter what the parameter
innodb_checksum_algorithm is assigned to.
For old files, the old checksum algorithms will continue to be
used. The value strict_full_crc32 will be equivalent to strict_crc32
and the value full_crc32 will be equivalent to crc32.
ROW_FORMAT=COMPRESSED tables will only use the old format.
These tables do not support new features, such as larger
innodb_page_size or instant ADD/DROP COLUMN. They may be
deprecated in the future. We do not want an unnecessary
file format change for them.
The new full_crc32() format also cleans up the MariaDB tablespace
flags. We will reserve flags to store the page_compressed
compression algorithm, and to store the compressed payload length,
so that checksum can be computed over the compressed (and
possibly encrypted) stream and can be validated without
decrypting or decompressing the page.
In the full_crc32 format, there no longer are separate before-encryption
and after-encryption checksums for pages. The single checksum is
computed on the page contents that is written to the file.
We do not make the new algorithm the default for two reasons.
First, MariaDB 10.4.2 was a beta release, and the default values
of parameters should not change after beta. Second, we did not
yet implement the full_crc32 format for page_compressed pages.
This will be fixed in MDEV-18644.
This is joint work with Marko Mäkelä.
2019-02-19 21:00:00 +02:00
|
|
|
@param[in] fsp_flags tablespace flags
|
2017-03-30 12:48:42 +02:00
|
|
|
@return whether the page is corrupted */
|
|
|
|
bool
|
2016-08-12 11:17:45 +03:00
|
|
|
buf_page_is_corrupted(
|
|
|
|
bool check_lsn,
|
|
|
|
const byte* read_buf,
|
MDEV-12026: Implement innodb_checksum_algorithm=full_crc32
MariaDB data-at-rest encryption (innodb_encrypt_tables)
had repurposed the same unused data field that was repurposed
in MySQL 5.7 (and MariaDB 10.2) for the Split Sequence Number (SSN)
field of SPATIAL INDEX. Because of this, MariaDB was unable to
support encryption on SPATIAL INDEX pages.
Furthermore, InnoDB page checksums skipped some bytes, and there
are multiple variations and checksum algorithms. By default,
InnoDB accepts all variations of all algorithms that ever existed.
This unnecessarily weakens the page checksums.
We hereby introduce two more innodb_checksum_algorithm variants
(full_crc32, strict_full_crc32) that are special in a way:
When either setting is active, newly created data files will
carry a flag (fil_space_t::full_crc32()) that indicates that
all pages of the file will use a full CRC-32C checksum over the
entire page contents (excluding the bytes where the checksum
is stored, at the very end of the page). Such files will always
use that checksum, no matter what the parameter
innodb_checksum_algorithm is assigned to.
For old files, the old checksum algorithms will continue to be
used. The value strict_full_crc32 will be equivalent to strict_crc32
and the value full_crc32 will be equivalent to crc32.
ROW_FORMAT=COMPRESSED tables will only use the old format.
These tables do not support new features, such as larger
innodb_page_size or instant ADD/DROP COLUMN. They may be
deprecated in the future. We do not want an unnecessary
file format change for them.
The new full_crc32() format also cleans up the MariaDB tablespace
flags. We will reserve flags to store the page_compressed
compression algorithm, and to store the compressed payload length,
so that checksum can be computed over the compressed (and
possibly encrypted) stream and can be validated without
decrypting or decompressing the page.
In the full_crc32 format, there no longer are separate before-encryption
and after-encryption checksums for pages. The single checksum is
computed on the page contents that is written to the file.
We do not make the new algorithm the default for two reasons.
First, MariaDB 10.4.2 was a beta release, and the default values
of parameters should not change after beta. Second, we did not
yet implement the full_crc32 format for page_compressed pages.
This will be fixed in MDEV-18644.
This is joint work with Marko Mäkelä.
2019-02-19 21:00:00 +02:00
|
|
|
ulint fsp_flags)
|
MDEV-11759: Encryption code in MariaDB 10.1/10.2 causes
compatibility problems
Pages that are encrypted contain post encryption checksum on
different location that normal checksum fields. Therefore,
we should before decryption check this checksum to avoid
unencrypting corrupted pages. After decryption we can use
traditional checksum check to detect if page is corrupted
or unencryption was done using incorrect key.
Pages that are page compressed do not contain any checksum,
here we need to fist unencrypt, decompress and finally
use tradional checksum check to detect page corruption
or that we used incorrect key in unencryption.
buf0buf.cc: buf_page_is_corrupted() mofified so that
compressed pages are skipped.
buf0buf.h, buf_block_init(), buf_page_init_low():
removed unnecessary page_encrypted, page_compressed,
stored_checksum, valculated_checksum fields from
buf_page_t
buf_page_get_gen(): use new buf_page_check_corrupt() function
to detect corrupted pages.
buf_page_check_corrupt(): If page was not yet decrypted
check if post encryption checksum still matches.
If page is not anymore encrypted, use buf_page_is_corrupted()
traditional checksum method.
If page is detected as corrupted and it is not encrypted
we print corruption message to error log.
If page is still encrypted or it was encrypted and now
corrupted, we will print message that page is
encrypted to error log.
buf_page_io_complete(): use new buf_page_check_corrupt()
function to detect corrupted pages.
buf_page_decrypt_after_read(): Verify post encryption
checksum before tring to decrypt.
fil0crypt.cc: fil_encrypt_buf() verify post encryption
checksum and ind fil_space_decrypt() return true
if we really decrypted the page.
fil_space_verify_crypt_checksum(): rewrite to use
the method used when calculating post encryption
checksum. We also check if post encryption checksum
matches that traditional checksum check does not
match.
fil0fil.ic: Add missed page type encrypted and page
compressed to fil_get_page_type_name()
Note that this change does not yet fix innochecksum tool,
that will be done in separate MDEV.
Fix test failures caused by buf page corruption injection.
2017-02-06 10:47:55 +02:00
|
|
|
MY_ATTRIBUTE((warn_unused_result));
|
2017-08-04 13:57:26 +03:00
|
|
|
|
2019-12-23 07:14:51 +02:00
|
|
|
inline void *aligned_malloc(size_t size, size_t align)
|
|
|
|
{
|
|
|
|
#ifdef _MSC_VER
|
|
|
|
return _aligned_malloc(size, align);
|
|
|
|
#else
|
|
|
|
void *result;
|
|
|
|
if (posix_memalign(&result, align, size))
|
|
|
|
result= NULL;
|
|
|
|
return result;
|
|
|
|
#endif
|
|
|
|
}
|
2017-08-03 08:29:36 +03:00
|
|
|
|
2019-12-23 07:14:51 +02:00
|
|
|
inline void aligned_free(void *ptr)
|
|
|
|
{
|
|
|
|
#ifdef _MSC_VER
|
|
|
|
_aligned_free(ptr);
|
|
|
|
#else
|
|
|
|
free(ptr);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
MDEV-12026: Implement innodb_checksum_algorithm=full_crc32
MariaDB data-at-rest encryption (innodb_encrypt_tables)
had repurposed the same unused data field that was repurposed
in MySQL 5.7 (and MariaDB 10.2) for the Split Sequence Number (SSN)
field of SPATIAL INDEX. Because of this, MariaDB was unable to
support encryption on SPATIAL INDEX pages.
Furthermore, InnoDB page checksums skipped some bytes, and there
are multiple variations and checksum algorithms. By default,
InnoDB accepts all variations of all algorithms that ever existed.
This unnecessarily weakens the page checksums.
We hereby introduce two more innodb_checksum_algorithm variants
(full_crc32, strict_full_crc32) that are special in a way:
When either setting is active, newly created data files will
carry a flag (fil_space_t::full_crc32()) that indicates that
all pages of the file will use a full CRC-32C checksum over the
entire page contents (excluding the bytes where the checksum
is stored, at the very end of the page). Such files will always
use that checksum, no matter what the parameter
innodb_checksum_algorithm is assigned to.
For old files, the old checksum algorithms will continue to be
used. The value strict_full_crc32 will be equivalent to strict_crc32
and the value full_crc32 will be equivalent to crc32.
ROW_FORMAT=COMPRESSED tables will only use the old format.
These tables do not support new features, such as larger
innodb_page_size or instant ADD/DROP COLUMN. They may be
deprecated in the future. We do not want an unnecessary
file format change for them.
The new full_crc32() format also cleans up the MariaDB tablespace
flags. We will reserve flags to store the page_compressed
compression algorithm, and to store the compressed payload length,
so that checksum can be computed over the compressed (and
possibly encrypted) stream and can be validated without
decrypting or decompressing the page.
In the full_crc32 format, there no longer are separate before-encryption
and after-encryption checksums for pages. The single checksum is
computed on the page contents that is written to the file.
We do not make the new algorithm the default for two reasons.
First, MariaDB 10.4.2 was a beta release, and the default values
of parameters should not change after beta. Second, we did not
yet implement the full_crc32 format for page_compressed pages.
This will be fixed in MDEV-18644.
This is joint work with Marko Mäkelä.
2019-02-19 21:00:00 +02:00
|
|
|
/** Read the key version from the page. In full crc32 format,
|
|
|
|
key version is stored at {0-3th} bytes. In other format, it is
|
|
|
|
stored in 26th position.
|
|
|
|
@param[in] read_buf database page
|
|
|
|
@param[in] fsp_flags tablespace flags
|
|
|
|
@return key version of the page. */
|
|
|
|
inline uint32_t buf_page_get_key_version(const byte* read_buf, ulint fsp_flags)
|
|
|
|
{
|
2020-05-07 12:25:00 +03:00
|
|
|
static_assert(FIL_PAGE_FCRC32_KEY_VERSION == 0, "compatibility");
|
|
|
|
return fil_space_t::full_crc32(fsp_flags)
|
|
|
|
? mach_read_from_4(my_assume_aligned<4>(read_buf))
|
|
|
|
: mach_read_from_4(my_assume_aligned<2>
|
|
|
|
(read_buf + FIL_PAGE_FILE_FLUSH_LSN_OR_KEY_VERSION));
|
MDEV-12026: Implement innodb_checksum_algorithm=full_crc32
MariaDB data-at-rest encryption (innodb_encrypt_tables)
had repurposed the same unused data field that was repurposed
in MySQL 5.7 (and MariaDB 10.2) for the Split Sequence Number (SSN)
field of SPATIAL INDEX. Because of this, MariaDB was unable to
support encryption on SPATIAL INDEX pages.
Furthermore, InnoDB page checksums skipped some bytes, and there
are multiple variations and checksum algorithms. By default,
InnoDB accepts all variations of all algorithms that ever existed.
This unnecessarily weakens the page checksums.
We hereby introduce two more innodb_checksum_algorithm variants
(full_crc32, strict_full_crc32) that are special in a way:
When either setting is active, newly created data files will
carry a flag (fil_space_t::full_crc32()) that indicates that
all pages of the file will use a full CRC-32C checksum over the
entire page contents (excluding the bytes where the checksum
is stored, at the very end of the page). Such files will always
use that checksum, no matter what the parameter
innodb_checksum_algorithm is assigned to.
For old files, the old checksum algorithms will continue to be
used. The value strict_full_crc32 will be equivalent to strict_crc32
and the value full_crc32 will be equivalent to crc32.
ROW_FORMAT=COMPRESSED tables will only use the old format.
These tables do not support new features, such as larger
innodb_page_size or instant ADD/DROP COLUMN. They may be
deprecated in the future. We do not want an unnecessary
file format change for them.
The new full_crc32() format also cleans up the MariaDB tablespace
flags. We will reserve flags to store the page_compressed
compression algorithm, and to store the compressed payload length,
so that checksum can be computed over the compressed (and
possibly encrypted) stream and can be validated without
decrypting or decompressing the page.
In the full_crc32 format, there no longer are separate before-encryption
and after-encryption checksums for pages. The single checksum is
computed on the page contents that is written to the file.
We do not make the new algorithm the default for two reasons.
First, MariaDB 10.4.2 was a beta release, and the default values
of parameters should not change after beta. Second, we did not
yet implement the full_crc32 format for page_compressed pages.
This will be fixed in MDEV-18644.
This is joint work with Marko Mäkelä.
2019-02-19 21:00:00 +02:00
|
|
|
}
|
|
|
|
|
MDEV-18644: Support full_crc32 for page_compressed
This is a follow-up task to MDEV-12026, which introduced
innodb_checksum_algorithm=full_crc32 and a simpler page format.
MDEV-12026 did not enable full_crc32 for page_compressed tables,
which we will be doing now.
This is joint work with Thirunarayanan Balathandayuthapani.
For innodb_checksum_algorithm=full_crc32 we change the
page_compressed format as follows:
FIL_PAGE_TYPE: The most significant bit will be set to indicate
page_compressed format. The least significant bits will contain
the compressed page size, rounded up to a multiple of 256 bytes.
The checksum will be stored in the last 4 bytes of the page
(whether it is the full page or a page_compressed page whose
size is determined by FIL_PAGE_TYPE), covering all preceding
bytes of the page. If encryption is used, then the page will
be encrypted between compression and computing the checksum.
For page_compressed, FIL_PAGE_LSN will not be repeated at
the end of the page.
FSP_SPACE_FLAGS (already implemented as part of MDEV-12026):
We will store the innodb_compression_algorithm that may be used
to compress pages. Previously, the choice of algorithm was written
to each compressed data page separately, and one would be unable
to know in advance which compression algorithm(s) are used.
fil_space_t::full_crc32_page_compressed_len(): Determine if the
page_compressed algorithm of the tablespace needs to know the
exact length of the compressed data. If yes, we will reserve and
write an extra byte for this right before the checksum.
buf_page_is_compressed(): Determine if a page uses page_compressed
(in any innodb_checksum_algorithm).
fil_page_decompress(): Pass also fil_space_t::flags so that the
format can be determined.
buf_page_is_zeroes(): Check if a page is full of zero bytes.
buf_page_full_crc32_is_corrupted(): Renamed from
buf_encrypted_full_crc32_page_is_corrupted(). For full_crc32,
we always simply validate the checksum to the page contents,
while the physical page size is explicitly specified by an
unencrypted part of the page header.
buf_page_full_crc32_size(): Determine the size of a full_crc32 page.
buf_dblwr_check_page_lsn(): Make this a debug-only function, because
it involves potentially costly lookups of fil_space_t.
create_table_info_t::check_table_options(),
ha_innobase::check_if_supported_inplace_alter(): Do allow the creation
of SPATIAL INDEX with full_crc32 also when page_compressed is used.
commit_cache_norebuild(): Preserve the compression algorithm when
updating the page_compression_level.
dict_tf_to_fsp_flags(): Set the flags for page compression algorithm.
FIXME: Maybe there should be a table option page_compression_algorithm
and a session variable to back it?
2019-03-18 14:08:43 +02:00
|
|
|
/** Read the compression info from the page. In full crc32 format,
|
|
|
|
compression info is at MSB of page type. In other format, it is
|
|
|
|
stored in page type.
|
|
|
|
@param[in] read_buf database page
|
|
|
|
@param[in] fsp_flags tablespace flags
|
|
|
|
@return true if page is compressed. */
|
|
|
|
inline bool buf_page_is_compressed(const byte* read_buf, ulint fsp_flags)
|
|
|
|
{
|
2020-05-07 17:15:34 +03:00
|
|
|
uint16_t page_type= fil_page_get_type(read_buf);
|
2020-05-07 12:25:00 +03:00
|
|
|
return fil_space_t::full_crc32(fsp_flags)
|
|
|
|
? !!(page_type & 1U << FIL_PAGE_COMPRESS_FCRC32_MARKER)
|
|
|
|
: page_type == FIL_PAGE_PAGE_COMPRESSED;
|
MDEV-18644: Support full_crc32 for page_compressed
This is a follow-up task to MDEV-12026, which introduced
innodb_checksum_algorithm=full_crc32 and a simpler page format.
MDEV-12026 did not enable full_crc32 for page_compressed tables,
which we will be doing now.
This is joint work with Thirunarayanan Balathandayuthapani.
For innodb_checksum_algorithm=full_crc32 we change the
page_compressed format as follows:
FIL_PAGE_TYPE: The most significant bit will be set to indicate
page_compressed format. The least significant bits will contain
the compressed page size, rounded up to a multiple of 256 bytes.
The checksum will be stored in the last 4 bytes of the page
(whether it is the full page or a page_compressed page whose
size is determined by FIL_PAGE_TYPE), covering all preceding
bytes of the page. If encryption is used, then the page will
be encrypted between compression and computing the checksum.
For page_compressed, FIL_PAGE_LSN will not be repeated at
the end of the page.
FSP_SPACE_FLAGS (already implemented as part of MDEV-12026):
We will store the innodb_compression_algorithm that may be used
to compress pages. Previously, the choice of algorithm was written
to each compressed data page separately, and one would be unable
to know in advance which compression algorithm(s) are used.
fil_space_t::full_crc32_page_compressed_len(): Determine if the
page_compressed algorithm of the tablespace needs to know the
exact length of the compressed data. If yes, we will reserve and
write an extra byte for this right before the checksum.
buf_page_is_compressed(): Determine if a page uses page_compressed
(in any innodb_checksum_algorithm).
fil_page_decompress(): Pass also fil_space_t::flags so that the
format can be determined.
buf_page_is_zeroes(): Check if a page is full of zero bytes.
buf_page_full_crc32_is_corrupted(): Renamed from
buf_encrypted_full_crc32_page_is_corrupted(). For full_crc32,
we always simply validate the checksum to the page contents,
while the physical page size is explicitly specified by an
unencrypted part of the page header.
buf_page_full_crc32_size(): Determine the size of a full_crc32 page.
buf_dblwr_check_page_lsn(): Make this a debug-only function, because
it involves potentially costly lookups of fil_space_t.
create_table_info_t::check_table_options(),
ha_innobase::check_if_supported_inplace_alter(): Do allow the creation
of SPATIAL INDEX with full_crc32 also when page_compressed is used.
commit_cache_norebuild(): Preserve the compression algorithm when
updating the page_compression_level.
dict_tf_to_fsp_flags(): Set the flags for page compression algorithm.
FIXME: Maybe there should be a table option page_compression_algorithm
and a session variable to back it?
2019-03-18 14:08:43 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/** Get the compressed or uncompressed size of a full_crc32 page.
|
|
|
|
@param[in] buf page_compressed or uncompressed page
|
|
|
|
@param[out] comp whether the page could be compressed
|
|
|
|
@param[out] cr whether the page could be corrupted
|
|
|
|
@return the payload size in the file page */
|
|
|
|
inline uint buf_page_full_crc32_size(const byte* buf, bool* comp, bool* cr)
|
|
|
|
{
|
2020-05-07 17:15:34 +03:00
|
|
|
uint t = fil_page_get_type(buf);
|
MDEV-18644: Support full_crc32 for page_compressed
This is a follow-up task to MDEV-12026, which introduced
innodb_checksum_algorithm=full_crc32 and a simpler page format.
MDEV-12026 did not enable full_crc32 for page_compressed tables,
which we will be doing now.
This is joint work with Thirunarayanan Balathandayuthapani.
For innodb_checksum_algorithm=full_crc32 we change the
page_compressed format as follows:
FIL_PAGE_TYPE: The most significant bit will be set to indicate
page_compressed format. The least significant bits will contain
the compressed page size, rounded up to a multiple of 256 bytes.
The checksum will be stored in the last 4 bytes of the page
(whether it is the full page or a page_compressed page whose
size is determined by FIL_PAGE_TYPE), covering all preceding
bytes of the page. If encryption is used, then the page will
be encrypted between compression and computing the checksum.
For page_compressed, FIL_PAGE_LSN will not be repeated at
the end of the page.
FSP_SPACE_FLAGS (already implemented as part of MDEV-12026):
We will store the innodb_compression_algorithm that may be used
to compress pages. Previously, the choice of algorithm was written
to each compressed data page separately, and one would be unable
to know in advance which compression algorithm(s) are used.
fil_space_t::full_crc32_page_compressed_len(): Determine if the
page_compressed algorithm of the tablespace needs to know the
exact length of the compressed data. If yes, we will reserve and
write an extra byte for this right before the checksum.
buf_page_is_compressed(): Determine if a page uses page_compressed
(in any innodb_checksum_algorithm).
fil_page_decompress(): Pass also fil_space_t::flags so that the
format can be determined.
buf_page_is_zeroes(): Check if a page is full of zero bytes.
buf_page_full_crc32_is_corrupted(): Renamed from
buf_encrypted_full_crc32_page_is_corrupted(). For full_crc32,
we always simply validate the checksum to the page contents,
while the physical page size is explicitly specified by an
unencrypted part of the page header.
buf_page_full_crc32_size(): Determine the size of a full_crc32 page.
buf_dblwr_check_page_lsn(): Make this a debug-only function, because
it involves potentially costly lookups of fil_space_t.
create_table_info_t::check_table_options(),
ha_innobase::check_if_supported_inplace_alter(): Do allow the creation
of SPATIAL INDEX with full_crc32 also when page_compressed is used.
commit_cache_norebuild(): Preserve the compression algorithm when
updating the page_compression_level.
dict_tf_to_fsp_flags(): Set the flags for page compression algorithm.
FIXME: Maybe there should be a table option page_compression_algorithm
and a session variable to back it?
2019-03-18 14:08:43 +02:00
|
|
|
uint page_size = uint(srv_page_size);
|
|
|
|
|
|
|
|
if (!(t & 1U << FIL_PAGE_COMPRESS_FCRC32_MARKER)) {
|
|
|
|
return page_size;
|
|
|
|
}
|
|
|
|
|
|
|
|
t &= ~(1U << FIL_PAGE_COMPRESS_FCRC32_MARKER);
|
|
|
|
t <<= 8;
|
|
|
|
|
|
|
|
if (t < page_size) {
|
|
|
|
page_size = t;
|
|
|
|
if (comp) {
|
|
|
|
*comp = true;
|
|
|
|
}
|
|
|
|
} else if (cr) {
|
|
|
|
*cr = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
return page_size;
|
|
|
|
}
|
|
|
|
|
2019-12-05 06:42:31 +02:00
|
|
|
#ifndef UNIV_INNOCHECKSUM
|
2017-09-06 14:01:15 +03:00
|
|
|
/** Dump a page to stderr.
|
|
|
|
@param[in] read_buf database page
|
2019-02-06 19:50:11 +02:00
|
|
|
@param[in] zip_size compressed page size, or 0 */
|
|
|
|
void buf_page_print(const byte* read_buf, ulint zip_size = 0)
|
2017-09-06 19:28:51 +03:00
|
|
|
ATTRIBUTE_COLD __attribute__((nonnull));
|
2014-02-26 19:11:54 +01:00
|
|
|
/********************************************************************//**
|
|
|
|
Decompress a block.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return TRUE if successful */
|
2014-02-26 19:11:54 +01:00
|
|
|
ibool
|
|
|
|
buf_zip_decompress(
|
|
|
|
/*===============*/
|
|
|
|
buf_block_t* block, /*!< in/out: block */
|
|
|
|
ibool check); /*!< in: TRUE=verify the page checksum */
|
2016-12-30 15:04:10 +02:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
#ifdef UNIV_DEBUG
|
2020-02-12 14:45:21 +02:00
|
|
|
/** @return the number of latched pages in the buffer pool */
|
|
|
|
ulint buf_get_latched_pages_number();
|
2014-02-26 19:11:54 +01:00
|
|
|
#endif /* UNIV_DEBUG */
|
|
|
|
/*********************************************************************//**
|
|
|
|
Prints info of the buffer i/o. */
|
|
|
|
void
|
|
|
|
buf_print_io(
|
|
|
|
/*=========*/
|
|
|
|
FILE* file); /*!< in: file where to print */
|
2020-02-12 14:45:21 +02:00
|
|
|
/** Collect buffer pool metadata.
|
|
|
|
@param[out] pool_info buffer pool metadata */
|
|
|
|
void buf_stats_get_pool_info(buf_pool_info_t *pool_info);
|
|
|
|
|
2015-05-26 10:01:12 +03:00
|
|
|
/** Refresh the statistics used to print per-second averages. */
|
2020-02-12 14:45:21 +02:00
|
|
|
void buf_refresh_io_stats();
|
|
|
|
|
|
|
|
/** Invalidate all pages in the buffer pool.
|
|
|
|
All pages must be in a replaceable state (not modified or latched). */
|
|
|
|
void buf_pool_invalidate();
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
/*========================================================================
|
|
|
|
--------------------------- LOWER LEVEL ROUTINES -------------------------
|
|
|
|
=========================================================================*/
|
|
|
|
|
2016-08-12 11:17:45 +03:00
|
|
|
#ifdef UNIV_DEBUG
|
2014-02-26 19:11:54 +01:00
|
|
|
/*********************************************************************//**
|
|
|
|
Adds latch level info for the rw-lock protecting the buffer frame. This
|
|
|
|
should be called in the debug version after a successful latching of a
|
|
|
|
page if we know the latching order level of the acquired latch. */
|
|
|
|
UNIV_INLINE
|
|
|
|
void
|
|
|
|
buf_block_dbg_add_level(
|
|
|
|
/*====================*/
|
|
|
|
buf_block_t* block, /*!< in: buffer page
|
|
|
|
where we have acquired latch */
|
2016-08-12 11:17:45 +03:00
|
|
|
latch_level_t level); /*!< in: latching order level */
|
|
|
|
#else /* UNIV_DEBUG */
|
2014-02-26 19:11:54 +01:00
|
|
|
# define buf_block_dbg_add_level(block, level) /* nothing */
|
2016-08-12 11:17:45 +03:00
|
|
|
#endif /* UNIV_DEBUG */
|
2016-12-30 15:04:10 +02:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
#ifdef UNIV_DEBUG
|
|
|
|
/*********************************************************************//**
|
|
|
|
Gets a pointer to the memory frame of a block.
|
2016-08-12 11:17:45 +03:00
|
|
|
@return pointer to the frame */
|
2014-02-26 19:11:54 +01:00
|
|
|
UNIV_INLINE
|
|
|
|
buf_frame_t*
|
|
|
|
buf_block_get_frame(
|
|
|
|
/*================*/
|
|
|
|
const buf_block_t* block) /*!< in: pointer to the control block */
|
2016-09-06 09:43:16 +03:00
|
|
|
MY_ATTRIBUTE((warn_unused_result));
|
2014-02-26 19:11:54 +01:00
|
|
|
#else /* UNIV_DEBUG */
|
|
|
|
# define buf_block_get_frame(block) (block)->frame
|
|
|
|
#endif /* UNIV_DEBUG */
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/*********************************************************************//**
|
|
|
|
Gets the compressed page descriptor corresponding to an uncompressed page
|
|
|
|
if applicable. */
|
|
|
|
#define buf_block_get_page_zip(block) \
|
2019-12-03 10:19:45 +02:00
|
|
|
(UNIV_LIKELY_NULL((block)->page.zip.data) ? &(block)->page.zip : NULL)
|
2019-08-08 22:53:33 +03:00
|
|
|
#define is_buf_block_get_page_zip(block) \
|
2019-12-03 10:19:45 +02:00
|
|
|
UNIV_LIKELY_NULL((block)->page.zip.data)
|
2016-09-06 09:43:16 +03:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Monitor the buffer page read/write activity, and increment corresponding
|
|
|
|
counter value in MONITOR_MODULE_BUF_PAGE.
|
|
|
|
@param bpage buffer page whose read or write was completed
|
|
|
|
@param io_type BUF_IO_READ or BUF_IO_WRITE */
|
|
|
|
ATTRIBUTE_COLD __attribute__((nonnull))
|
|
|
|
void buf_page_monitor(const buf_page_t *bpage, buf_io_fix io_type);
|
2016-08-12 11:17:45 +03:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Complete a read request of a file page to buf_pool.
|
|
|
|
@param bpage recently read page
|
|
|
|
@param node data file
|
2017-05-05 10:25:29 +03:00
|
|
|
@return whether the operation succeeded
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
@retval DB_SUCCESS always when writing, or if a read page was OK
|
|
|
|
@retval DB_PAGE_CORRUPTED if the checksum fails on a page read
|
|
|
|
@retval DB_DECRYPTION_FAILED if the page cannot be decrypted */
|
|
|
|
dberr_t buf_page_read_complete(buf_page_t *bpage, const fil_node_t &node);
|
2016-08-12 11:17:45 +03:00
|
|
|
|
|
|
|
/** Calculate aligned buffer pool size based on srv_buf_pool_chunk_unit,
|
|
|
|
if needed.
|
|
|
|
@param[in] size size in bytes
|
|
|
|
@return aligned size */
|
|
|
|
UNIV_INLINE
|
|
|
|
ulint
|
|
|
|
buf_pool_size_align(
|
|
|
|
ulint size);
|
|
|
|
|
MDEV-12026: Implement innodb_checksum_algorithm=full_crc32
MariaDB data-at-rest encryption (innodb_encrypt_tables)
had repurposed the same unused data field that was repurposed
in MySQL 5.7 (and MariaDB 10.2) for the Split Sequence Number (SSN)
field of SPATIAL INDEX. Because of this, MariaDB was unable to
support encryption on SPATIAL INDEX pages.
Furthermore, InnoDB page checksums skipped some bytes, and there
are multiple variations and checksum algorithms. By default,
InnoDB accepts all variations of all algorithms that ever existed.
This unnecessarily weakens the page checksums.
We hereby introduce two more innodb_checksum_algorithm variants
(full_crc32, strict_full_crc32) that are special in a way:
When either setting is active, newly created data files will
carry a flag (fil_space_t::full_crc32()) that indicates that
all pages of the file will use a full CRC-32C checksum over the
entire page contents (excluding the bytes where the checksum
is stored, at the very end of the page). Such files will always
use that checksum, no matter what the parameter
innodb_checksum_algorithm is assigned to.
For old files, the old checksum algorithms will continue to be
used. The value strict_full_crc32 will be equivalent to strict_crc32
and the value full_crc32 will be equivalent to crc32.
ROW_FORMAT=COMPRESSED tables will only use the old format.
These tables do not support new features, such as larger
innodb_page_size or instant ADD/DROP COLUMN. They may be
deprecated in the future. We do not want an unnecessary
file format change for them.
The new full_crc32() format also cleans up the MariaDB tablespace
flags. We will reserve flags to store the page_compressed
compression algorithm, and to store the compressed payload length,
so that checksum can be computed over the compressed (and
possibly encrypted) stream and can be validated without
decrypting or decompressing the page.
In the full_crc32 format, there no longer are separate before-encryption
and after-encryption checksums for pages. The single checksum is
computed on the page contents that is written to the file.
We do not make the new algorithm the default for two reasons.
First, MariaDB 10.4.2 was a beta release, and the default values
of parameters should not change after beta. Second, we did not
yet implement the full_crc32 format for page_compressed pages.
This will be fixed in MDEV-18644.
This is joint work with Marko Mäkelä.
2019-02-19 21:00:00 +02:00
|
|
|
/** Verify that post encryption checksum match with the calculated checksum.
|
|
|
|
This function should be called only if tablespace contains crypt data metadata.
|
|
|
|
@param[in] page page frame
|
|
|
|
@param[in] fsp_flags tablespace flags
|
|
|
|
@return true if page is encrypted and OK, false otherwise */
|
|
|
|
bool buf_page_verify_crypt_checksum(
|
|
|
|
const byte* page,
|
|
|
|
ulint fsp_flags);
|
|
|
|
|
MDEV-21132 Remove buf_page_t::newest_modification
At each mini-transaction commit, the log sequence number of the
mini-transaction must be written to each modified page, so that
it will be available in the FIL_PAGE_LSN field when the page is
being read in crash recovery.
InnoDB was unnecessarily allocating redundant storage for the
field, in buf_page_t::newest_modification. Let us access
FIL_PAGE_LSN directly.
Furthermore, on ALTER TABLE...IMPORT TABLESPACE, let us write
0 to FIL_PAGE_LSN instead of using log_sys.lsn.
buf_flush_init_for_writing(), buf_flush_update_zip_checksum(),
fil_encrypt_buf_for_full_crc32(), fil_encrypt_buf(),
fil_space_encrypt(): Remove the parameter lsn.
buf_page_get_newest_modification(): Merge with the only caller.
buf_tmp_reserve_compression_buf(), buf_tmp_page_encrypt(),
buf_page_encrypt(): Define static in the same compilation unit
with the only caller.
PageConverter::m_current_lsn: Remove. Write 0 to FIL_PAGE_LSN
on ALTER TABLE...IMPORT TABLESPACE.
2019-11-25 09:39:51 +02:00
|
|
|
/** Calculate a ROW_FORMAT=COMPRESSED page checksum and update the page.
|
|
|
|
@param[in,out] page page to update
|
|
|
|
@param[in] size compressed page size */
|
|
|
|
void buf_flush_update_zip_checksum(buf_frame_t* page, ulint size);
|
2014-12-22 16:53:17 +02:00
|
|
|
|
2015-04-01 19:37:00 +03:00
|
|
|
/** @brief The temporary memory structure.
|
2014-12-22 16:53:17 +02:00
|
|
|
|
2015-04-01 19:37:00 +03:00
|
|
|
NOTE! The definition appears here only for other modules of this
|
|
|
|
directory (buf) to see it. Do not use from outside! */
|
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
class buf_tmp_buffer_t
|
|
|
|
{
|
|
|
|
/** whether this slot is reserved */
|
|
|
|
std::atomic<bool> reserved;
|
2018-06-13 16:15:21 +03:00
|
|
|
public:
|
2020-03-18 21:48:00 +02:00
|
|
|
/** For encryption, the data needs to be copied to a separate buffer
|
|
|
|
before it's encrypted&written. The buffer block itself can be replaced
|
|
|
|
while a write of crypt_buf to file is in progress. */
|
|
|
|
byte *crypt_buf;
|
|
|
|
/** buffer for fil_page_compress(), for flushing page_compressed pages */
|
|
|
|
byte *comp_buf;
|
|
|
|
/** pointer to resulting buffer after encryption or compression;
|
|
|
|
not separately allocated memory */
|
|
|
|
byte *out_buf;
|
|
|
|
|
|
|
|
/** Release the slot */
|
|
|
|
void release() { reserved.store(false, std::memory_order_relaxed); }
|
|
|
|
|
|
|
|
/** Acquire the slot
|
|
|
|
@return whether the slot was acquired */
|
|
|
|
bool acquire() { return !reserved.exchange(true, std::memory_order_relaxed);}
|
|
|
|
|
|
|
|
/** Allocate a buffer for encryption, decryption or decompression. */
|
|
|
|
void allocate()
|
|
|
|
{
|
|
|
|
if (!crypt_buf)
|
|
|
|
crypt_buf= static_cast<byte*>
|
|
|
|
(aligned_malloc(srv_page_size, srv_page_size));
|
|
|
|
}
|
2018-10-13 23:22:05 +04:00
|
|
|
};
|
2014-12-22 16:53:17 +02:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/** The common buffer control block structure
|
|
|
|
for compressed and uncompressed frames */
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
class buf_pool_t;
|
2014-05-05 18:20:28 +02:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
class buf_page_t
|
|
|
|
{
|
|
|
|
friend buf_pool_t;
|
|
|
|
friend buf_block_t;
|
|
|
|
/** @name General fields */
|
|
|
|
/* @{ */
|
|
|
|
|
|
|
|
public: // FIXME: fix fil_iterate()
|
|
|
|
/** Page id. Protected by buf_pool.hash_lock_get(id) when
|
|
|
|
the page is in buf_pool.page_hash. */
|
|
|
|
page_id_t id_;
|
|
|
|
private:
|
|
|
|
/** Count of how manyfold this block is currently bufferfixed. */
|
|
|
|
Atomic_counter<uint32_t> buf_fix_count_;
|
|
|
|
|
2021-06-23 13:13:11 +03:00
|
|
|
/** log sequence number of the START of the log entry written of the
|
|
|
|
oldest modification to this block which has not yet been written
|
|
|
|
to the data file;
|
|
|
|
|
|
|
|
0 if no modifications are pending;
|
|
|
|
1 if no modifications are pending, but the block is in buf_pool.flush_list;
|
|
|
|
2 if modifications are pending, but the block is not in buf_pool.flush_list
|
|
|
|
(because id().space() is the temporary tablespace). */
|
2021-06-26 11:16:40 +03:00
|
|
|
Atomic_relaxed<lsn_t> oldest_modification_;
|
2021-06-23 13:13:11 +03:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** type of pending I/O operation; protected by buf_pool.mutex
|
|
|
|
if in_LRU_list */
|
|
|
|
Atomic_relaxed<buf_io_fix> io_fix_;
|
|
|
|
/** Block state. @see in_file().
|
|
|
|
State transitions between in_file() states and to
|
|
|
|
BUF_BLOCK_REMOVE_HASH are protected by buf_pool.hash_lock_get(id)
|
|
|
|
when the block is in buf_pool.page_hash.
|
|
|
|
Other transitions when in_LRU_list are protected by buf_pool.mutex. */
|
|
|
|
buf_page_state state_;
|
2014-05-05 18:20:28 +02:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
public:
|
|
|
|
/** buf_pool.page_hash link; protected by buf_pool.hash_lock_get(id) */
|
|
|
|
buf_page_t *hash;
|
|
|
|
/* @} */
|
2014-02-26 19:11:54 +01:00
|
|
|
page_zip_des_t zip; /*!< compressed page; zip.data
|
|
|
|
(but not the data it points to) is
|
2020-03-18 21:48:00 +02:00
|
|
|
also protected by buf_pool.mutex;
|
2014-02-26 19:11:54 +01:00
|
|
|
state == BUF_BLOCK_ZIP_PAGE and
|
|
|
|
zip.data == NULL means an active
|
2020-03-18 21:48:00 +02:00
|
|
|
buf_pool.watch */
|
2013-12-19 14:36:38 +02:00
|
|
|
|
2015-04-01 19:37:00 +03:00
|
|
|
buf_tmp_buffer_t* slot; /*!< Slot for temporary memory
|
|
|
|
used for encryption/compression
|
|
|
|
or NULL */
|
2014-02-26 19:11:54 +01:00
|
|
|
#ifdef UNIV_DEBUG
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** whether this->list is in buf_pool.zip_hash; protected by buf_pool.mutex */
|
|
|
|
bool in_zip_hash;
|
|
|
|
/** whether this->LRU is in buf_pool.LRU (in_file() holds);
|
|
|
|
protected by buf_pool.mutex */
|
|
|
|
bool in_LRU_list;
|
|
|
|
/** whether this is in buf_pool.page_hash (in_file() holds);
|
|
|
|
protected by buf_pool.mutex */
|
|
|
|
bool in_page_hash;
|
|
|
|
/** whether this->list is in buf_pool.free (state() == BUF_BLOCK_NOT_USED);
|
|
|
|
protected by buf_pool.flush_list_mutex */
|
|
|
|
bool in_free_list;
|
2014-02-26 19:11:54 +01:00
|
|
|
#endif /* UNIV_DEBUG */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** list member in one of the lists of buf_pool; protected by
|
|
|
|
buf_pool.mutex or buf_pool.flush_list_mutex
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
state() == BUF_BLOCK_NOT_USED: buf_pool.free or buf_pool.withdraw
|
2014-02-26 19:11:54 +01:00
|
|
|
|
2020-10-15 09:49:20 +03:00
|
|
|
in_file() && oldest_modification():
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
buf_pool.flush_list (protected by buf_pool.flush_list_mutex)
|
2014-02-26 19:11:54 +01:00
|
|
|
|
2020-10-15 09:49:20 +03:00
|
|
|
The contents is undefined if in_file() && !oldest_modification(),
|
|
|
|
or if state() is BUF_BLOCK_MEMORY or BUF_BLOCK_REMOVE_HASH. */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
UT_LIST_NODE_T(buf_page_t) list;
|
|
|
|
|
|
|
|
/** @name LRU replacement algorithm fields.
|
|
|
|
Protected by buf_pool.mutex. */
|
2014-02-26 19:11:54 +01:00
|
|
|
/* @{ */
|
|
|
|
|
|
|
|
UT_LIST_NODE_T(buf_page_t) LRU;
|
|
|
|
/*!< node of the LRU list */
|
|
|
|
unsigned old:1; /*!< TRUE if the block is in the old
|
2020-03-18 21:48:00 +02:00
|
|
|
blocks in buf_pool.LRU_old */
|
2014-02-26 19:11:54 +01:00
|
|
|
unsigned freed_page_clock:31;/*!< the value of
|
2020-03-18 21:48:00 +02:00
|
|
|
buf_pool.freed_page_clock
|
2014-02-26 19:11:54 +01:00
|
|
|
when this block was the last
|
|
|
|
time put to the head of the
|
|
|
|
LRU list; a thread is allowed
|
|
|
|
to read this for heuristic
|
|
|
|
purposes without holding any
|
|
|
|
mutex or latch */
|
|
|
|
/* @} */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
Atomic_counter<unsigned> access_time; /*!< time of first access, or
|
2014-02-26 19:11:54 +01:00
|
|
|
0 if the block was never accessed
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
in the buffer pool.
|
2020-02-06 09:00:19 +02:00
|
|
|
|
|
|
|
For state==BUF_BLOCK_MEMORY
|
|
|
|
blocks, this field can be repurposed
|
|
|
|
for something else.
|
|
|
|
|
|
|
|
When this field counts log records
|
|
|
|
and bytes allocated for recv_sys.pages,
|
|
|
|
the field is protected by
|
|
|
|
recv_sys_t::mutex. */
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
/** Change buffer entries for the page exist.
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
Protected by io_fix()==BUF_IO_READ or by buf_block_t::lock. */
|
MDEV-19514 Defer change buffer merge until pages are requested
We will remove the InnoDB background operation of merging buffered
changes to secondary index leaf pages. Changes will only be merged as a
result of an operation that accesses a secondary index leaf page,
such as a SQL statement that performs a lookup via that index,
or is modifying the index. Also ROLLBACK and some background operations,
such as purging the history of committed transactions, or computing
index cardinality statistics, can cause change buffer merge.
Encryption key rotation will not perform change buffer merge.
The motivation of this change is to simplify the I/O logic and to
allow crash recovery to happen in the background (MDEV-14481).
We also hope that this will reduce the number of "mystery" crashes
due to corrupted data. Because change buffer merge will typically
take place as a result of executing SQL statements, there should be
a clearer connection between the crash and the SQL statements that
were executed when the server crashed.
In many cases, a slight performance improvement was observed.
This is joint work with Thirunarayanan Balathandayuthapani
and was tested by Axel Schwenke and Matthias Leich.
The InnoDB monitor counter innodb_ibuf_merge_usec will be removed.
On slow shutdown (innodb_fast_shutdown=0), we will continue to
merge all buffered changes (and purge all undo log history).
Two InnoDB configuration parameters will be changed as follows:
innodb_disable_background_merge: Removed.
This parameter existed only in debug builds.
All change buffer merges will use synchronous reads.
innodb_force_recovery will be changed as follows:
* innodb_force_recovery=4 will be the same as innodb_force_recovery=3
(the change buffer merge cannot be disabled; it can only happen as
a result of an operation that accesses a secondary index leaf page).
The option used to be capable of corrupting secondary index leaf pages.
Now that capability is removed, and innodb_force_recovery=4 becomes 'safe'.
* innodb_force_recovery=5 (which essentially hard-wires
SET GLOBAL TRANSACTION ISOLATION LEVEL READ UNCOMMITTED)
becomes safe to use. Bogus data can be returned to SQL, but
persistent InnoDB data files will not be corrupted further.
* innodb_force_recovery=6 (ignore the redo log files)
will be the only option that can potentially cause
persistent corruption of InnoDB data files.
Code changes:
buf_page_t::ibuf_exist: New flag, to indicate whether buffered
changes exist for a buffer pool page. Pages with pending changes
can be returned by buf_page_get_gen(). Previously, the changes
were always merged inside buf_page_get_gen() if needed.
ibuf_page_exists(const buf_page_t&): Check if a buffered changes
exist for an X-latched or read-fixed page.
buf_page_get_gen(): Add the parameter allow_ibuf_merge=false.
All callers that know that they may be accessing a secondary index
leaf page must pass this parameter as allow_ibuf_merge=true,
unless it does not matter for that caller whether all buffered
changes have been applied. Assert that whenever allow_ibuf_merge
holds, the page actually is a leaf page. Attempt change buffer
merge only to secondary B-tree index leaf pages.
btr_block_get(): Add parameter 'bool merge'.
All callers of btr_block_get() should know whether the page could be
a secondary index leaf page. If it is not, we should avoid consulting
the change buffer bitmap to even consider a merge. This is the main
interface to requesting index pages from the buffer pool.
ibuf_merge_or_delete_for_page(), recv_recover_page(): Replace
buf_page_get_known_nowait() with much simpler logic, because
it is now guaranteed that that the block is x-latched or read-fixed.
mlog_init_t::mark_ibuf_exist(): Renamed from mlog_init_t::ibuf_merge().
On crash recovery, we will no longer merge any buffered changes
for the pages that we read into the buffer pool during the last batch
of applying log records.
buf_page_get_gen_known_nowait(), BUF_MAKE_YOUNG, BUF_KEEP_OLD: Remove.
btr_search_guess_on_hash(): Merge buf_page_get_gen_known_nowait()
to its only remaining caller.
buf_page_make_young_if_needed(): Define as an inline function.
Add the parameter buf_pool.
buf_page_peek_if_young(), buf_page_peek_if_too_old(): Add the
parameter buf_pool.
fil_space_validate_for_mtr_commit(): Remove a bogus comment
about background merge of the change buffer.
btr_cur_open_at_rnd_pos_func(), btr_cur_search_to_nth_level_func(),
btr_cur_open_at_index_side_func(): Use narrower data types and scopes.
ibuf_read_merge_pages(): Replaces buf_read_ibuf_merge_pages().
Merge the change buffer by invoking buf_page_get_gen().
2019-10-11 17:28:15 +03:00
|
|
|
bool ibuf_exist;
|
2019-01-07 01:18:19 +04:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Block initialization status. Can be modified while holding io_fix()
|
MDEV-15528 Punch holes when pages are freed
When a InnoDB data file page is freed, its contents becomes garbage,
and any storage allocated in the data file is wasted. During flushing,
InnoDB initializes the page with zeros if scrubbing is enabled. If the
tablespace is compressed then InnoDB should punch a hole else ignore the
flushing of the freed page.
buf_page_t:
- Replaced the variable file_page_was_freed, init_on_flush in buf_page_t
with status enum variable.
- Changed all debug assert of file_page_was_freed to DBUG_ASSERT
of buf_page_t::status
Removed buf_page_set_file_page_was_freed(),
buf_page_reset_file_page_was_freed().
buf_page_free(): Newly added function which takes X-lock on the page
before marking the status as FREED. So that InnoDB flush handler can
avoid concurrent flush of the freed page. Also while flushing the page,
InnoDB make sure that redo log which does freeing of the page also written
to the disk. Currently, this function only marks the page as FREED if
it is in buffer pool
buf_flush_freed_page(): Newly added function which initializes zeros
asynchorously if innodb_immediate_scrub_data_uncompressed is enabled.
Punch a hole to the file synchorously if page_compressed is enabled.
Reset the io_fix to NORMAL. Release the block from flush list and
associated mutex before writing zeros or punch a hole to the file.
buf_flush_page(): Removed the unnecessary usage of temporary
variable "flush"
fil_io(): Introduce new parameter called punch_hole. It allows fil_io()
to punch the hole to the file for the given offset.
buf_page_create(): Let the callers assign buf_page_t::status.
Every caller should eventually invoke mtr_t::init().
fsp_page_create(): Remove the unused mtr_t parameter.
In all other callers of buf_page_create() except fsp_page_create(),
before invoking mtr_t::init(), invoke
mtr_t::sx_latch_at_savepoint() or mtr_t::x_latch_at_savepoint().
mtr_t::init(): Initialize buf_page_t::status also for the temporary
tablespace (when redo logging is disabled), to avoid assertion failures.
2020-03-09 13:25:33 +05:30
|
|
|
or buf_block_t::lock X-latch */
|
|
|
|
enum {
|
|
|
|
/** the page was read normally and should be flushed normally */
|
|
|
|
NORMAL = 0,
|
|
|
|
/** the page was (re)initialized, and the doublewrite buffer can be
|
|
|
|
skipped on the next flush */
|
|
|
|
INIT_ON_FLUSH,
|
|
|
|
/** the page was freed and need to be flushed.
|
|
|
|
For page_compressed, page flush will punch a hole to free space.
|
|
|
|
Else if innodb_immediate_scrub_data_uncompressed, the page will
|
|
|
|
be overwritten with zeroes. */
|
|
|
|
FREED
|
|
|
|
} status;
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
buf_page_t() : id_(0)
|
|
|
|
{
|
|
|
|
static_assert(BUF_BLOCK_NOT_USED == 0, "compatibility");
|
|
|
|
memset((void*) this, 0, sizeof *this);
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Initialize some fields */
|
|
|
|
void init()
|
|
|
|
{
|
|
|
|
io_fix_= BUF_IO_NONE;
|
|
|
|
buf_fix_count_= 0;
|
|
|
|
old= 0;
|
|
|
|
freed_page_clock= 0;
|
|
|
|
access_time= 0;
|
|
|
|
oldest_modification_= 0;
|
|
|
|
slot= nullptr;
|
|
|
|
ibuf_exist= false;
|
|
|
|
status= NORMAL;
|
|
|
|
ut_d(in_zip_hash= false);
|
|
|
|
ut_d(in_free_list= false);
|
|
|
|
ut_d(in_LRU_list= false);
|
|
|
|
ut_d(in_page_hash= false);
|
|
|
|
HASH_INVALIDATE(this, hash);
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Initialize some more fields */
|
|
|
|
void init(buf_page_state state, page_id_t id, uint32_t buf_fix_count= 0)
|
|
|
|
{
|
|
|
|
init();
|
|
|
|
state_= state;
|
|
|
|
id_= id;
|
|
|
|
buf_fix_count_= buf_fix_count;
|
|
|
|
}
|
|
|
|
|
2020-06-11 17:30:33 +03:00
|
|
|
/** Initialize some more fields */
|
|
|
|
void init(page_id_t id, uint32_t buf_fix_count= 0)
|
|
|
|
{
|
|
|
|
init();
|
|
|
|
id_= id;
|
|
|
|
buf_fix_count_= buf_fix_count;
|
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
public:
|
|
|
|
const page_id_t &id() const { return id_; }
|
|
|
|
buf_page_state state() const { return state_; }
|
|
|
|
uint32_t buf_fix_count() const { return buf_fix_count_; }
|
|
|
|
buf_io_fix io_fix() const { return io_fix_; }
|
|
|
|
void io_unfix()
|
|
|
|
{
|
|
|
|
ut_d(const auto old_io_fix= io_fix());
|
|
|
|
ut_ad(old_io_fix == BUF_IO_READ || old_io_fix == BUF_IO_PIN);
|
|
|
|
io_fix_= BUF_IO_NONE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** @return if this belongs to buf_pool.unzip_LRU */
|
|
|
|
bool belongs_to_unzip_LRU() const
|
|
|
|
{
|
2020-06-11 17:30:33 +03:00
|
|
|
return zip.data && state() != BUF_BLOCK_ZIP_PAGE;
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
inline void add_buf_fix_count(uint32_t count);
|
|
|
|
inline void set_buf_fix_count(uint32_t count);
|
|
|
|
inline void set_state(buf_page_state state);
|
|
|
|
inline void set_io_fix(buf_io_fix io_fix);
|
|
|
|
inline void set_corrupt_id();
|
|
|
|
|
2021-06-23 13:13:11 +03:00
|
|
|
/** @return the log sequence number of the oldest pending modification
|
2021-06-26 11:16:40 +03:00
|
|
|
@retval 0 if the block is being removed from (or not in) buf_pool.flush_list
|
2021-06-23 13:13:11 +03:00
|
|
|
@retval 1 if the block is in buf_pool.flush_list but not modified
|
|
|
|
@retval 2 if the block belongs to the temporary tablespace and
|
|
|
|
has unwritten changes */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
lsn_t oldest_modification() const { return oldest_modification_; }
|
2021-06-26 11:16:40 +03:00
|
|
|
/** @return the log sequence number of the oldest pending modification,
|
|
|
|
@retval 0 if the block is definitely not in buf_pool.flush_list
|
|
|
|
@retval 1 if the block is in buf_pool.flush_list but not modified
|
|
|
|
@retval 2 if the block belongs to the temporary tablespace and
|
|
|
|
has unwritten changes */
|
|
|
|
lsn_t oldest_modification_acquire() const
|
|
|
|
{ return oldest_modification_.load(std::memory_order_acquire); }
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Set oldest_modification when adding to buf_pool.flush_list */
|
|
|
|
inline void set_oldest_modification(lsn_t lsn);
|
2021-06-24 21:55:10 +03:00
|
|
|
/** Clear oldest_modification after removing from buf_pool.flush_list */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline void clear_oldest_modification();
|
2021-06-23 13:13:11 +03:00
|
|
|
/** Note that a block is no longer dirty, while not removing
|
|
|
|
it from buf_pool.flush_list */
|
|
|
|
inline void clear_oldest_modification(bool temporary);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
|
2020-12-09 09:22:13 +02:00
|
|
|
/** Notify that a page in a temporary tablespace has been modified. */
|
|
|
|
void set_temp_modified()
|
|
|
|
{
|
|
|
|
ut_ad(fsp_is_system_temporary(id().space()));
|
|
|
|
ut_ad(state() == BUF_BLOCK_FILE_PAGE);
|
|
|
|
ut_ad(!oldest_modification());
|
2021-06-23 13:13:11 +03:00
|
|
|
oldest_modification_= 2;
|
2020-12-09 09:22:13 +02:00
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Prepare to release a file page to buf_pool.free. */
|
|
|
|
void free_file_page()
|
|
|
|
{
|
|
|
|
ut_ad(state() == BUF_BLOCK_REMOVE_HASH);
|
2021-06-26 11:16:40 +03:00
|
|
|
/* buf_LRU_block_free_non_file_page() asserts !oldest_modification() */
|
|
|
|
ut_d(oldest_modification_= 0;)
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
set_corrupt_id();
|
|
|
|
ut_d(set_state(BUF_BLOCK_MEMORY));
|
|
|
|
}
|
|
|
|
|
|
|
|
void fix() { buf_fix_count_++; }
|
2019-01-07 01:18:19 +04:00
|
|
|
uint32_t unfix()
|
|
|
|
{
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
uint32_t count= buf_fix_count_--;
|
2019-01-07 01:18:19 +04:00
|
|
|
ut_ad(count != 0);
|
|
|
|
return count - 1;
|
|
|
|
}
|
2019-02-06 19:50:11 +02:00
|
|
|
|
|
|
|
/** @return the physical size, in bytes */
|
|
|
|
ulint physical_size() const
|
|
|
|
{
|
|
|
|
return zip.ssize ? (UNIV_ZIP_SIZE_MIN >> 1) << zip.ssize : srv_page_size;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** @return the ROW_FORMAT=COMPRESSED physical size, in bytes
|
|
|
|
@retval 0 if not compressed */
|
|
|
|
ulint zip_size() const
|
|
|
|
{
|
|
|
|
return zip.ssize ? (UNIV_ZIP_SIZE_MIN >> 1) << zip.ssize : 0;
|
|
|
|
}
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
|
2020-10-26 15:59:30 +02:00
|
|
|
/** @return the byte offset of the page within a file */
|
|
|
|
os_offset_t physical_offset() const
|
|
|
|
{
|
|
|
|
os_offset_t o= id().page_no();
|
|
|
|
return zip.ssize
|
|
|
|
? o << (zip.ssize + (UNIV_ZIP_SIZE_SHIFT_MIN - 1))
|
|
|
|
: o << srv_page_size_shift;
|
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** @return whether the block is mapped to a data file */
|
|
|
|
bool in_file() const
|
|
|
|
{
|
|
|
|
switch (state_) {
|
|
|
|
case BUF_BLOCK_ZIP_PAGE:
|
|
|
|
case BUF_BLOCK_FILE_PAGE:
|
|
|
|
return true;
|
|
|
|
case BUF_BLOCK_NOT_USED:
|
|
|
|
case BUF_BLOCK_MEMORY:
|
|
|
|
case BUF_BLOCK_REMOVE_HASH:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
ut_error;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** @return whether the block is modified and ready for flushing */
|
|
|
|
inline bool ready_for_flush() const;
|
|
|
|
/** @return whether the state can be changed to BUF_BLOCK_NOT_USED */
|
|
|
|
bool ready_for_replace() const
|
|
|
|
{ return !oldest_modification() && can_relocate(); }
|
|
|
|
/** @return whether the block can be relocated in memory.
|
|
|
|
The block can be dirty, but it must not be I/O-fixed or bufferfixed. */
|
|
|
|
inline bool can_relocate() const;
|
|
|
|
/** @return whether the block has been flagged old in buf_pool.LRU */
|
|
|
|
inline bool is_old() const;
|
|
|
|
/** Set whether a block is old in buf_pool.LRU */
|
|
|
|
inline void set_old(bool old);
|
|
|
|
/** Flag a page accessed in buf_pool
|
|
|
|
@return whether this is not the first access */
|
|
|
|
bool set_accessed()
|
|
|
|
{
|
|
|
|
if (is_accessed()) return true;
|
|
|
|
access_time= static_cast<uint32_t>(ut_time_ms());
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
/** @return ut_time_ms() at the time of first access of a block in buf_pool
|
|
|
|
@retval 0 if not accessed */
|
|
|
|
unsigned is_accessed() const { ut_ad(in_file()); return access_time; }
|
2014-02-26 19:11:54 +01:00
|
|
|
};
|
|
|
|
|
|
|
|
/** The buffer control block structure */
|
|
|
|
|
|
|
|
struct buf_block_t{
|
|
|
|
|
|
|
|
/** @name General fields */
|
|
|
|
/* @{ */
|
|
|
|
|
|
|
|
buf_page_t page; /*!< page information; this must
|
|
|
|
be the first field, so that
|
2020-03-18 21:48:00 +02:00
|
|
|
buf_pool.page_hash can point
|
2014-02-26 19:11:54 +01:00
|
|
|
to buf_page_t or buf_block_t */
|
|
|
|
byte* frame; /*!< pointer to buffer frame which
|
2018-04-27 13:49:25 +03:00
|
|
|
is of size srv_page_size, and
|
2014-02-26 19:11:54 +01:00
|
|
|
aligned to an address divisible by
|
2018-04-27 13:49:25 +03:00
|
|
|
srv_page_size */
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
rw_lock_t lock; /*!< read-write lock of the buffer
|
2016-08-12 11:17:45 +03:00
|
|
|
frame */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
#ifdef UNIV_DEBUG
|
|
|
|
/** whether page.list is in buf_pool.withdraw
|
|
|
|
((state() == BUF_BLOCK_NOT_USED)) and the buffer pool is being shrunk;
|
|
|
|
protected by buf_pool.mutex */
|
|
|
|
bool in_withdraw_list;
|
|
|
|
/** whether unzip_LRU is in buf_pool.unzip_LRU
|
|
|
|
(state() == BUF_BLOCK_FILE_PAGE and zip.data != nullptr);
|
|
|
|
protected by buf_pool.mutex */
|
|
|
|
bool in_unzip_LRU_list;
|
|
|
|
#endif
|
2014-02-26 19:11:54 +01:00
|
|
|
UT_LIST_NODE_T(buf_block_t) unzip_LRU;
|
|
|
|
/*!< node of the decompressed LRU list;
|
|
|
|
a block is in the unzip_LRU list
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
if page.state() == BUF_BLOCK_FILE_PAGE
|
2014-02-26 19:11:54 +01:00
|
|
|
and page.zip.data != NULL */
|
|
|
|
/* @} */
|
|
|
|
/** @name Optimistic search field */
|
|
|
|
/* @{ */
|
|
|
|
|
|
|
|
ib_uint64_t modify_clock; /*!< this clock is incremented every
|
|
|
|
time a pointer to a record on the
|
|
|
|
page may become obsolete; this is
|
|
|
|
used in the optimistic cursor
|
|
|
|
positioning: if the modify clock has
|
|
|
|
not changed, we know that the pointer
|
|
|
|
is still valid; this field may be
|
|
|
|
changed if the thread (1) owns the
|
|
|
|
pool mutex and the page is not
|
|
|
|
bufferfixed, or (2) the thread has an
|
|
|
|
x-latch on the block */
|
|
|
|
/* @} */
|
2017-02-23 23:05:12 +02:00
|
|
|
#ifdef BTR_CUR_HASH_ADAPT
|
2014-02-26 19:11:54 +01:00
|
|
|
/** @name Hash search fields (unprotected)
|
|
|
|
NOTE that these fields are NOT protected by any semaphore! */
|
|
|
|
/* @{ */
|
|
|
|
|
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC
The -Wconversion in GCC seems to be stricter than in clang.
GCC at least since version 4.4.7 issues truncation warnings for
assignments to bitfields, while clang 10 appears to only issue
warnings when the sizes in bytes rounded to the nearest integer
powers of 2 are different.
Before GCC 10.0.0, -Wconversion required more casts and would not
allow some operations, such as x<<=1 or x+=1 on a data type that
is narrower than int.
GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining
about x|=y even when x and y are compatible types that are narrower
than int. Hence, we must rewrite some x|=y as
x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion.
In GCC 6 and later, the warning for assigning wider to bitfields
that are narrower than 8, 16, or 32 bits can be suppressed by
applying a bitwise & with the exact bitmask of the bitfield.
For older GCC, we must disable -Wconversion for GCC 4 or 5 in such
cases.
The bitwise negation operator appears to promote short integers
to a wider type, and hence we must add explicit truncation casts
around them. Microsoft Visual C does not allow a static_cast to
truncate a constant, such as static_cast<byte>(1) truncating int.
Hence, we will use the constructor-style cast byte(~1) for such cases.
This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0,
clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019)
on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
2020-03-12 19:46:41 +02:00
|
|
|
volatile uint16_t n_bytes; /*!< recommended prefix length for hash
|
2016-08-12 11:17:45 +03:00
|
|
|
search: number of bytes in
|
|
|
|
an incomplete last field */
|
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC
The -Wconversion in GCC seems to be stricter than in clang.
GCC at least since version 4.4.7 issues truncation warnings for
assignments to bitfields, while clang 10 appears to only issue
warnings when the sizes in bytes rounded to the nearest integer
powers of 2 are different.
Before GCC 10.0.0, -Wconversion required more casts and would not
allow some operations, such as x<<=1 or x+=1 on a data type that
is narrower than int.
GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining
about x|=y even when x and y are compatible types that are narrower
than int. Hence, we must rewrite some x|=y as
x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion.
In GCC 6 and later, the warning for assigning wider to bitfields
that are narrower than 8, 16, or 32 bits can be suppressed by
applying a bitwise & with the exact bitmask of the bitfield.
For older GCC, we must disable -Wconversion for GCC 4 or 5 in such
cases.
The bitwise negation operator appears to promote short integers
to a wider type, and hence we must add explicit truncation casts
around them. Microsoft Visual C does not allow a static_cast to
truncate a constant, such as static_cast<byte>(1) truncating int.
Hence, we will use the constructor-style cast byte(~1) for such cases.
This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0,
clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019)
on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
2020-03-12 19:46:41 +02:00
|
|
|
volatile uint16_t n_fields; /*!< recommended prefix length for hash
|
2014-02-26 19:11:54 +01:00
|
|
|
search: number of full fields */
|
MDEV-21907: InnoDB: Enable -Wconversion on clang and GCC
The -Wconversion in GCC seems to be stricter than in clang.
GCC at least since version 4.4.7 issues truncation warnings for
assignments to bitfields, while clang 10 appears to only issue
warnings when the sizes in bytes rounded to the nearest integer
powers of 2 are different.
Before GCC 10.0.0, -Wconversion required more casts and would not
allow some operations, such as x<<=1 or x+=1 on a data type that
is narrower than int.
GCC 5 (but not GCC 4, GCC 6, or any later version) is complaining
about x|=y even when x and y are compatible types that are narrower
than int. Hence, we must rewrite some x|=y as
x=static_cast<byte>(x|y) or similar, or we must disable -Wconversion.
In GCC 6 and later, the warning for assigning wider to bitfields
that are narrower than 8, 16, or 32 bits can be suppressed by
applying a bitwise & with the exact bitmask of the bitfield.
For older GCC, we must disable -Wconversion for GCC 4 or 5 in such
cases.
The bitwise negation operator appears to promote short integers
to a wider type, and hence we must add explicit truncation casts
around them. Microsoft Visual C does not allow a static_cast to
truncate a constant, such as static_cast<byte>(1) truncating int.
Hence, we will use the constructor-style cast byte(~1) for such cases.
This has been tested at least with GCC 4.8.5, 5.4.0, 7.4.0, 9.2.1, 10.0.0,
clang 9.0.1, 10.0.0, and MSVC 14.22.27905 (Microsoft Visual Studio 2019)
on 64-bit and 32-bit targets (IA-32, AMD64, POWER 8, POWER 9, ARMv8).
2020-03-12 19:46:41 +02:00
|
|
|
uint16_t n_hash_helps; /*!< counter which controls building
|
|
|
|
of a new hash index for the page */
|
2016-08-12 11:17:45 +03:00
|
|
|
volatile bool left_side; /*!< true or false, depending on
|
2014-02-26 19:11:54 +01:00
|
|
|
whether the leftmost record of several
|
|
|
|
records with the same prefix should be
|
|
|
|
indexed in the hash index */
|
|
|
|
/* @} */
|
|
|
|
|
|
|
|
/** @name Hash search fields
|
2016-09-02 17:28:54 +03:00
|
|
|
These 5 fields may only be modified when:
|
|
|
|
we are holding the appropriate x-latch in btr_search_latches[], and
|
|
|
|
one of the following holds:
|
|
|
|
(1) the block state is BUF_BLOCK_FILE_PAGE, and
|
|
|
|
we are holding an s-latch or x-latch on buf_block_t::lock, or
|
|
|
|
(2) buf_block_t::buf_fix_count == 0, or
|
|
|
|
(3) the block state is BUF_BLOCK_REMOVE_HASH.
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
An exception to this is when we init or create a page
|
|
|
|
in the buffer pool in buf0buf.cc.
|
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
Another exception for buf_pool_t::clear_hash_index() is that
|
2016-09-02 17:28:54 +03:00
|
|
|
assigning block->index = NULL (and block->n_pointers = 0)
|
|
|
|
is allowed whenever btr_search_own_all(RW_LOCK_X).
|
|
|
|
|
2020-06-18 12:17:37 +03:00
|
|
|
Another exception is that ha_insert_for_fold() may
|
2016-09-02 17:28:54 +03:00
|
|
|
decrement n_pointers without holding the appropriate latch
|
|
|
|
in btr_search_latches[]. Thus, n_pointers must be
|
|
|
|
protected by atomic memory access.
|
|
|
|
|
|
|
|
This implies that the fields may be read without race
|
|
|
|
condition whenever any of the following hold:
|
|
|
|
- the btr_search_latches[] s-latch or x-latch is being held, or
|
|
|
|
- the block state is not BUF_BLOCK_FILE_PAGE or BUF_BLOCK_REMOVE_HASH,
|
|
|
|
and holding some latch prevents the state from changing to that.
|
|
|
|
|
|
|
|
Some use of assert_block_ahi_empty() or assert_block_ahi_valid()
|
2020-03-18 21:48:00 +02:00
|
|
|
is prone to race conditions while buf_pool_t::clear_hash_index() is
|
2016-09-02 17:28:54 +03:00
|
|
|
executing (the adaptive hash index is being disabled). Such use
|
|
|
|
is explicitly commented. */
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
/* @{ */
|
|
|
|
|
2017-02-23 23:05:12 +02:00
|
|
|
# if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
|
2018-12-28 17:53:50 +04:00
|
|
|
Atomic_counter<ulint>
|
|
|
|
n_pointers; /*!< used in debugging: the number of
|
2014-02-26 19:11:54 +01:00
|
|
|
pointers in the adaptive hash index
|
2016-09-02 17:28:54 +03:00
|
|
|
pointing to this frame;
|
|
|
|
protected by atomic memory access
|
|
|
|
or btr_search_own_all(). */
|
|
|
|
# define assert_block_ahi_empty(block) \
|
2018-12-28 17:53:50 +04:00
|
|
|
ut_a((block)->n_pointers == 0)
|
2016-09-06 14:57:16 +03:00
|
|
|
# define assert_block_ahi_empty_on_init(block) do { \
|
MDEV-20377: Make WITH_MSAN more usable
MemorySanitizer (clang -fsanitize=memory) requires that all code
be compiled with instrumentation enabled. The only exception is the
C runtime library. Failure to use instrumented libraries will cause
bogus messages about memory being uninitialized.
In WITH_MSAN builds, we must avoid calling getservbyname(),
because even though it is a standard library function, it is
not instrumented, not even in clang 10.
Note: Before MariaDB Server 10.5, ./mtr will typically fail
due to the old PCRE library, which was updated in MDEV-14024.
The following cmake options were tested on 10.5
in commit 94d0bb4dbeb28a94d1f87fdd55f4297ff3df0157:
cmake \
-DCMAKE_C_FLAGS='-march=native -O2' \
-DCMAKE_CXX_FLAGS='-stdlib=libc++ -march=native -O2' \
-DWITH_EMBEDDED_SERVER=OFF -DWITH_UNIT_TESTS=OFF -DCMAKE_BUILD_TYPE=Debug \
-DWITH_INNODB_{BZIP2,LZ4,LZMA,LZO,SNAPPY}=OFF \
-DPLUGIN_{ARCHIVE,TOKUDB,MROONGA,OQGRAPH,ROCKSDB,CONNECT,SPIDER}=NO \
-DWITH_SAFEMALLOC=OFF \
-DWITH_{ZLIB,SSL,PCRE}=bundled \
-DHAVE_LIBAIO_H=0 \
-DWITH_MSAN=ON
MEM_MAKE_DEFINED(): An alias for VALGRIND_MAKE_MEM_DEFINED()
and __msan_unpoison().
MEM_GET_VBITS(), MEM_SET_VBITS(): Aliases for
VALGRIND_GET_VBITS(), VALGRIND_SET_VBITS(), __msan_copy_shadow().
InnoDB: Replace the UNIV_MEM_ macros with corresponding MEM_ macros.
ut_crc32_8_hw(), ut_crc32_64_low_hw(): Use the compiler built-in
functions instead of inline assembler when building WITH_MSAN.
This will require at least -msse4.2 when building for IA-32 or AMD64.
The inline assembler would not be instrumented, and would thus cause
bogus failures.
2020-07-01 17:23:00 +03:00
|
|
|
MEM_MAKE_DEFINED(&(block)->n_pointers, sizeof (block)->n_pointers); \
|
2016-09-06 14:57:16 +03:00
|
|
|
assert_block_ahi_empty(block); \
|
|
|
|
} while (0)
|
2016-09-02 17:28:54 +03:00
|
|
|
# define assert_block_ahi_valid(block) \
|
2018-12-28 17:53:50 +04:00
|
|
|
ut_a((block)->index || (block)->n_pointers == 0)
|
2016-09-02 17:28:54 +03:00
|
|
|
# else /* UNIV_AHI_DEBUG || UNIV_DEBUG */
|
|
|
|
# define assert_block_ahi_empty(block) /* nothing */
|
2016-09-06 14:57:16 +03:00
|
|
|
# define assert_block_ahi_empty_on_init(block) /* nothing */
|
2016-09-02 17:28:54 +03:00
|
|
|
# define assert_block_ahi_valid(block) /* nothing */
|
2017-02-23 23:05:12 +02:00
|
|
|
# endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
|
2014-02-26 19:11:54 +01:00
|
|
|
unsigned curr_n_fields:10;/*!< prefix length for hash indexing:
|
|
|
|
number of full fields */
|
|
|
|
unsigned curr_n_bytes:15;/*!< number of bytes in hash
|
|
|
|
indexing */
|
|
|
|
unsigned curr_left_side:1;/*!< TRUE or FALSE in hash indexing */
|
|
|
|
dict_index_t* index; /*!< Index for which the
|
|
|
|
adaptive hash index has been
|
|
|
|
created, or NULL if the page
|
|
|
|
does not exist in the
|
|
|
|
index. Note that it does not
|
|
|
|
guarantee that the index is
|
|
|
|
complete, though: there may
|
|
|
|
have been hash collisions,
|
|
|
|
record deletions, etc. */
|
2016-09-02 17:28:54 +03:00
|
|
|
/* @} */
|
|
|
|
#else /* BTR_CUR_HASH_ADAPT */
|
|
|
|
# define assert_block_ahi_empty(block) /* nothing */
|
2016-09-06 14:57:16 +03:00
|
|
|
# define assert_block_ahi_empty_on_init(block) /* nothing */
|
2016-09-02 17:28:54 +03:00
|
|
|
# define assert_block_ahi_valid(block) /* nothing */
|
2017-02-23 23:05:12 +02:00
|
|
|
#endif /* BTR_CUR_HASH_ADAPT */
|
2016-08-12 11:17:45 +03:00
|
|
|
# ifdef UNIV_DEBUG
|
2014-02-26 19:11:54 +01:00
|
|
|
/** @name Debug fields */
|
|
|
|
/* @{ */
|
2018-12-28 09:56:46 +02:00
|
|
|
rw_lock_t* debug_latch; /*!< in the debug version, each thread
|
2014-02-26 19:11:54 +01:00
|
|
|
which bufferfixes the block acquires
|
|
|
|
an s-latch here; so we can use the
|
|
|
|
debug utilities in sync0rw */
|
|
|
|
/* @} */
|
|
|
|
# endif
|
2019-01-07 01:18:19 +04:00
|
|
|
void fix() { page.fix(); }
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
uint32_t unfix()
|
|
|
|
{
|
2020-07-24 20:17:43 +05:30
|
|
|
ut_ad(page.buf_fix_count() || page.io_fix() != BUF_IO_NONE ||
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
page.state() == BUF_BLOCK_ZIP_PAGE ||
|
|
|
|
!rw_lock_own_flagged(&lock, RW_LOCK_FLAG_X | RW_LOCK_FLAG_S |
|
|
|
|
RW_LOCK_FLAG_SX));
|
2020-07-24 20:17:43 +05:30
|
|
|
return page.unfix();
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
}
|
2019-02-06 19:50:11 +02:00
|
|
|
|
|
|
|
/** @return the physical size, in bytes */
|
|
|
|
ulint physical_size() const { return page.physical_size(); }
|
|
|
|
|
|
|
|
/** @return the ROW_FORMAT=COMPRESSED physical size, in bytes
|
|
|
|
@retval 0 if not compressed */
|
|
|
|
ulint zip_size() const { return page.zip_size(); }
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Initialize the block.
|
2020-06-11 17:30:33 +03:00
|
|
|
@param page_id page identifier
|
|
|
|
@param zip_size ROW_FORMAT=COMPRESSED page size, or 0
|
|
|
|
@param fix initial buf_fix_count() */
|
|
|
|
void initialise(const page_id_t page_id, ulint zip_size, uint32_t fix= 0);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
};
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/**********************************************************************//**
|
2020-03-18 21:48:00 +02:00
|
|
|
Compute the hash fold value for blocks in buf_pool.zip_hash. */
|
2014-02-26 19:11:54 +01:00
|
|
|
/* @{ */
|
2018-04-27 14:26:43 +03:00
|
|
|
#define BUF_POOL_ZIP_FOLD_PTR(ptr) (ulint(ptr) >> srv_page_size_shift)
|
2014-02-26 19:11:54 +01:00
|
|
|
#define BUF_POOL_ZIP_FOLD(b) BUF_POOL_ZIP_FOLD_PTR((b)->frame)
|
|
|
|
#define BUF_POOL_ZIP_FOLD_BPAGE(b) BUF_POOL_ZIP_FOLD((buf_block_t*) (b))
|
|
|
|
/* @} */
|
|
|
|
|
2014-11-06 13:17:11 +02:00
|
|
|
/** A "Hazard Pointer" class used to iterate over page lists
|
|
|
|
inside the buffer pool. A hazard pointer is a buf_page_t pointer
|
|
|
|
which we intend to iterate over next and we want it remain valid
|
|
|
|
even after we release the buffer pool mutex. */
|
2020-02-12 14:45:21 +02:00
|
|
|
class HazardPointer
|
|
|
|
{
|
2014-11-06 13:17:11 +02:00
|
|
|
public:
|
2020-03-18 21:48:00 +02:00
|
|
|
virtual ~HazardPointer() {}
|
2014-11-06 13:17:11 +02:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/** @return current value */
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
buf_page_t *get() const { mysql_mutex_assert_owner(m_mutex); return m_hp; }
|
2014-11-06 13:17:11 +02:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Set current value
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
@param bpage buffer block to be set as hp */
|
2020-03-18 21:48:00 +02:00
|
|
|
void set(buf_page_t *bpage)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(m_mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(!bpage || bpage->in_file());
|
2020-03-18 21:48:00 +02:00
|
|
|
m_hp= bpage;
|
|
|
|
}
|
2014-11-06 13:17:11 +02:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Checks if a bpage is the hp
|
|
|
|
@param bpage buffer block to be compared
|
|
|
|
@return true if it is hp */
|
|
|
|
bool is_hp(const buf_page_t *bpage) const
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
{ mysql_mutex_assert_owner(m_mutex); return bpage == m_hp; }
|
2014-11-06 13:17:11 +02:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Adjust the value of hp. This happens when some
|
|
|
|
other thread working on the same list attempts to
|
|
|
|
remove the hp from the list. */
|
|
|
|
virtual void adjust(const buf_page_t*) = 0;
|
2014-11-06 13:17:11 +02:00
|
|
|
|
2016-08-12 11:17:45 +03:00
|
|
|
#ifdef UNIV_DEBUG
|
2020-03-18 21:48:00 +02:00
|
|
|
/** mutex that protects access to the m_hp. */
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
const mysql_mutex_t *m_mutex= nullptr;
|
2014-11-06 13:17:11 +02:00
|
|
|
#endif /* UNIV_DEBUG */
|
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
protected:
|
|
|
|
/** hazard pointer */
|
|
|
|
buf_page_t *m_hp= nullptr;
|
2014-11-06 13:17:11 +02:00
|
|
|
};
|
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Class implementing buf_pool.flush_list hazard pointer */
|
|
|
|
class FlushHp : public HazardPointer
|
|
|
|
{
|
2014-11-06 13:17:11 +02:00
|
|
|
public:
|
2020-03-18 21:48:00 +02:00
|
|
|
~FlushHp() override {}
|
|
|
|
|
|
|
|
/** Adjust the value of hp. This happens when some
|
|
|
|
other thread working on the same list attempts to
|
|
|
|
remove the hp from the list.
|
|
|
|
@param bpage buffer block to be compared */
|
|
|
|
void adjust(const buf_page_t *bpage) override
|
|
|
|
{
|
|
|
|
ut_ad(bpage != NULL);
|
2014-11-06 13:17:11 +02:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/* We only support reverse traversal for now. */
|
|
|
|
if (is_hp(bpage))
|
|
|
|
m_hp= UT_LIST_GET_PREV(list, m_hp);
|
2014-11-06 13:17:11 +02:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(!m_hp || m_hp->oldest_modification());
|
2020-03-18 21:48:00 +02:00
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
/** Class implementing buf_pool.LRU hazard pointer */
|
|
|
|
class LRUHp : public HazardPointer {
|
2014-11-06 13:17:11 +02:00
|
|
|
public:
|
2020-03-18 21:48:00 +02:00
|
|
|
~LRUHp() override {}
|
|
|
|
|
|
|
|
/** Adjust the value of hp. This happens when some
|
|
|
|
other thread working on the same list attempts to
|
|
|
|
remove the hp from the list.
|
|
|
|
@param bpage buffer block to be compared */
|
|
|
|
void adjust(const buf_page_t *bpage) override
|
|
|
|
{
|
|
|
|
ut_ad(bpage);
|
|
|
|
/** We only support reverse traversal for now. */
|
|
|
|
if (is_hp(bpage))
|
|
|
|
m_hp= UT_LIST_GET_PREV(LRU, m_hp);
|
|
|
|
|
|
|
|
ut_ad(!m_hp || m_hp->in_LRU_list);
|
|
|
|
}
|
2014-11-06 13:17:11 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
/** Special purpose iterators to be used when scanning the LRU list.
|
|
|
|
The idea is that when one thread finishes the scan it leaves the
|
|
|
|
itr in that position and the other thread can start scan from
|
|
|
|
there */
|
2020-03-18 21:48:00 +02:00
|
|
|
class LRUItr : public LRUHp {
|
2014-11-06 13:17:11 +02:00
|
|
|
public:
|
2020-03-18 21:48:00 +02:00
|
|
|
LRUItr() : LRUHp() {}
|
|
|
|
~LRUItr() override {}
|
|
|
|
|
|
|
|
/** Select from where to start a scan. If we have scanned
|
|
|
|
too deep into the LRU list it resets the value to the tail
|
|
|
|
of the LRU list.
|
|
|
|
@return buf_page_t from where to start scan. */
|
|
|
|
inline buf_page_t *start();
|
2014-11-06 13:17:11 +02:00
|
|
|
};
|
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/** Struct that is embedded in the free zip blocks */
|
|
|
|
struct buf_buddy_free_t {
|
|
|
|
union {
|
|
|
|
ulint size; /*!< size of the block */
|
|
|
|
byte bytes[FIL_PAGE_DATA];
|
|
|
|
/*!< stamp[FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID]
|
|
|
|
== BUF_BUDDY_FREE_STAMP denotes a free
|
|
|
|
block. If the space_id field of buddy
|
|
|
|
block != BUF_BUDDY_FREE_STAMP, the block
|
|
|
|
is not in any zip_free list. If the
|
|
|
|
space_id is BUF_BUDDY_FREE_STAMP then
|
|
|
|
stamp[0] will contain the
|
|
|
|
buddy block size. */
|
|
|
|
} stamp;
|
|
|
|
|
|
|
|
buf_page_t bpage; /*!< Embedded bpage descriptor */
|
|
|
|
UT_LIST_NODE_T(buf_buddy_free_t) list;
|
|
|
|
/*!< Node of zip_free list */
|
|
|
|
};
|
|
|
|
|
|
|
|
/** @brief The buffer pool statistics structure. */
|
|
|
|
struct buf_pool_stat_t{
|
|
|
|
ulint n_page_gets; /*!< number of page gets performed;
|
|
|
|
also successful searches through
|
|
|
|
the adaptive hash index are
|
|
|
|
counted as page gets; this field
|
|
|
|
is NOT protected by the buffer
|
|
|
|
pool mutex */
|
|
|
|
ulint n_pages_read; /*!< number read operations */
|
|
|
|
ulint n_pages_written;/*!< number write operations */
|
|
|
|
ulint n_pages_created;/*!< number of pages created
|
|
|
|
in the pool with no read */
|
|
|
|
ulint n_ra_pages_read_rnd;/*!< number of pages read in
|
|
|
|
as part of random read ahead */
|
|
|
|
ulint n_ra_pages_read;/*!< number of pages read in
|
|
|
|
as part of read ahead */
|
|
|
|
ulint n_ra_pages_evicted;/*!< number of read ahead
|
|
|
|
pages that are evicted without
|
|
|
|
being accessed */
|
|
|
|
ulint n_pages_made_young; /*!< number of pages made young, in
|
2020-06-08 13:14:27 +03:00
|
|
|
buf_page_make_young() */
|
2014-02-26 19:11:54 +01:00
|
|
|
ulint n_pages_not_made_young; /*!< number of pages not made
|
|
|
|
young because the first access
|
|
|
|
was not long enough ago, in
|
|
|
|
buf_page_peek_if_too_old() */
|
2021-01-07 11:18:13 +02:00
|
|
|
/** number of waits for eviction; writes protected by buf_pool.mutex */
|
|
|
|
ulint LRU_waits;
|
2014-02-26 19:11:54 +01:00
|
|
|
ulint LRU_bytes; /*!< LRU size in bytes */
|
|
|
|
ulint flush_list_bytes;/*!< flush_list size in bytes */
|
|
|
|
};
|
|
|
|
|
|
|
|
/** Statistics of buddy blocks of a given size. */
|
|
|
|
struct buf_buddy_stat_t {
|
|
|
|
/** Number of blocks allocated from the buddy system. */
|
|
|
|
ulint used;
|
|
|
|
/** Number of blocks relocated by the buddy system. */
|
|
|
|
ib_uint64_t relocated;
|
|
|
|
/** Total duration of block relocations, in microseconds. */
|
|
|
|
ib_uint64_t relocated_usec;
|
|
|
|
};
|
|
|
|
|
2020-02-12 14:45:21 +02:00
|
|
|
/** The buffer pool */
|
2020-03-18 21:48:00 +02:00
|
|
|
class buf_pool_t
|
2020-02-12 14:45:21 +02:00
|
|
|
{
|
2020-03-18 21:48:00 +02:00
|
|
|
/** A chunk of buffers */
|
|
|
|
struct chunk_t
|
|
|
|
{
|
|
|
|
/** number of elements in blocks[] */
|
|
|
|
size_t size;
|
|
|
|
/** memory allocated for the page frames */
|
|
|
|
unsigned char *mem;
|
|
|
|
/** descriptor of mem */
|
|
|
|
ut_new_pfx_t mem_pfx;
|
|
|
|
/** array of buffer control blocks */
|
|
|
|
buf_block_t *blocks;
|
|
|
|
|
|
|
|
/** Map of first page frame address to chunks[] */
|
|
|
|
using map= std::map<const void*, chunk_t*, std::less<const void*>,
|
|
|
|
ut_allocator<std::pair<const void* const,chunk_t*>>>;
|
|
|
|
/** Chunk map that may be under construction by buf_resize_thread() */
|
|
|
|
static map *map_reg;
|
|
|
|
/** Current chunk map for lookup only */
|
|
|
|
static map *map_ref;
|
|
|
|
|
|
|
|
/** @return the memory size bytes. */
|
|
|
|
size_t mem_size() const { return mem_pfx.m_size; }
|
|
|
|
|
|
|
|
/** Register the chunk */
|
|
|
|
void reg() { map_reg->emplace(map::value_type(blocks->frame, this)); }
|
|
|
|
|
|
|
|
/** Allocate a chunk of buffer frames.
|
|
|
|
@param bytes requested size
|
|
|
|
@return whether the allocation succeeded */
|
|
|
|
inline bool create(size_t bytes);
|
|
|
|
|
|
|
|
#ifdef UNIV_DEBUG
|
|
|
|
/** Find a block that points to a ROW_FORMAT=COMPRESSED page
|
|
|
|
@param data pointer to the start of a ROW_FORMAT=COMPRESSED page frame
|
|
|
|
@return the block
|
|
|
|
@retval nullptr if not found */
|
|
|
|
const buf_block_t *contains_zip(const void *data) const
|
|
|
|
{
|
|
|
|
const buf_block_t *block= blocks;
|
|
|
|
for (auto i= size; i--; block++)
|
|
|
|
if (block->page.zip.data == data)
|
|
|
|
return block;
|
|
|
|
return nullptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Check that all blocks are in a replaceable state.
|
|
|
|
@return address of a non-free block
|
|
|
|
@retval nullptr if all freed */
|
|
|
|
inline const buf_block_t *not_freed() const;
|
|
|
|
#endif /* UNIV_DEBUG */
|
|
|
|
};
|
|
|
|
|
|
|
|
/** Withdraw blocks from the buffer pool until meeting withdraw_target.
|
|
|
|
@return whether retry is needed */
|
|
|
|
inline bool withdraw_blocks();
|
|
|
|
|
|
|
|
/** Determine if a pointer belongs to a buf_block_t. It can be a pointer to
|
|
|
|
the buf_block_t itself or a member of it.
|
|
|
|
@param ptr a pointer that will not be dereferenced
|
|
|
|
@return whether the ptr belongs to a buf_block_t struct */
|
|
|
|
bool is_block_field(const void *ptr) const
|
|
|
|
{
|
|
|
|
const chunk_t *chunk= chunks;
|
|
|
|
const chunk_t *const echunk= chunk + ut_min(n_chunks, n_chunks_new);
|
|
|
|
|
|
|
|
/* TODO: protect chunks with a mutex (the older pointer will
|
|
|
|
currently remain during resize()) */
|
|
|
|
for (; chunk < echunk; chunk++)
|
|
|
|
if (ptr >= reinterpret_cast<const void*>(chunk->blocks) &&
|
|
|
|
ptr < reinterpret_cast<const void*>(chunk->blocks + chunk->size))
|
|
|
|
return true;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Try to reallocate a control block.
|
|
|
|
@param block control block to reallocate
|
|
|
|
@return whether the reallocation succeeded */
|
|
|
|
inline bool realloc(buf_block_t *block);
|
|
|
|
|
|
|
|
public:
|
|
|
|
bool is_initialised() const { return chunks != nullptr; }
|
|
|
|
|
|
|
|
/** Create the buffer pool.
|
|
|
|
@return whether the creation failed */
|
|
|
|
bool create();
|
|
|
|
|
|
|
|
/** Clean up after successful create() */
|
|
|
|
void close();
|
|
|
|
|
|
|
|
/** Resize from srv_buf_pool_old_size to srv_buf_pool_size. */
|
|
|
|
inline void resize();
|
|
|
|
|
|
|
|
/** @return whether resize() is in progress */
|
|
|
|
bool resize_in_progress() const
|
|
|
|
{
|
2020-10-30 11:15:30 +02:00
|
|
|
return UNIV_UNLIKELY(resizing.load(std::memory_order_relaxed));
|
2020-03-18 21:48:00 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/** @return the current size in blocks */
|
|
|
|
size_t get_n_pages() const
|
|
|
|
{
|
|
|
|
ut_ad(is_initialised());
|
|
|
|
size_t size= 0;
|
|
|
|
for (auto j= n_chunks; j--; )
|
|
|
|
size+= chunks[j].size;
|
|
|
|
return size;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Determine whether a frame is intended to be withdrawn during resize().
|
|
|
|
@param ptr pointer within a buf_block_t::frame
|
|
|
|
@return whether the frame will be withdrawn */
|
|
|
|
bool will_be_withdrawn(const byte *ptr) const
|
|
|
|
{
|
|
|
|
ut_ad(curr_size < old_size);
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
#ifdef SAFE_MUTEX
|
|
|
|
if (resizing.load(std::memory_order_relaxed))
|
|
|
|
mysql_mutex_assert_owner(&mutex);
|
|
|
|
#endif /* SAFE_MUTEX */
|
2020-03-18 21:48:00 +02:00
|
|
|
|
|
|
|
for (const chunk_t *chunk= chunks + n_chunks_new,
|
|
|
|
* const echunk= chunks + n_chunks;
|
|
|
|
chunk != echunk; chunk++)
|
|
|
|
if (ptr >= chunk->blocks->frame &&
|
|
|
|
ptr < (chunk->blocks + chunk->size - 1)->frame + srv_page_size)
|
|
|
|
return true;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Determine whether a block is intended to be withdrawn during resize().
|
|
|
|
@param bpage buffer pool block
|
|
|
|
@return whether the frame will be withdrawn */
|
|
|
|
bool will_be_withdrawn(const buf_page_t &bpage) const
|
|
|
|
{
|
|
|
|
ut_ad(curr_size < old_size);
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
#ifdef SAFE_MUTEX
|
|
|
|
if (resizing.load(std::memory_order_relaxed))
|
|
|
|
mysql_mutex_assert_owner(&mutex);
|
|
|
|
#endif /* SAFE_MUTEX */
|
2020-03-18 21:48:00 +02:00
|
|
|
|
|
|
|
for (const chunk_t *chunk= chunks + n_chunks_new,
|
|
|
|
* const echunk= chunks + n_chunks;
|
|
|
|
chunk != echunk; chunk++)
|
|
|
|
if (&bpage >= &chunk->blocks->page &&
|
|
|
|
&bpage < &chunk->blocks[chunk->size].page)
|
|
|
|
return true;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Release and evict a corrupted page.
|
|
|
|
@param bpage page that was being read */
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
ATTRIBUTE_COLD void corrupted_evict(buf_page_t *bpage);
|
|
|
|
|
|
|
|
/** Release a memory block to the buffer pool. */
|
|
|
|
ATTRIBUTE_COLD void free_block(buf_block_t *block);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
#ifdef UNIV_DEBUG
|
|
|
|
/** Find a block that points to a ROW_FORMAT=COMPRESSED page
|
|
|
|
@param data pointer to the start of a ROW_FORMAT=COMPRESSED page frame
|
|
|
|
@return the block
|
|
|
|
@retval nullptr if not found */
|
|
|
|
const buf_block_t *contains_zip(const void *data) const
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&mutex);
|
2020-03-18 21:48:00 +02:00
|
|
|
for (const chunk_t *chunk= chunks, * const end= chunks + n_chunks;
|
|
|
|
chunk != end; chunk++)
|
|
|
|
if (const buf_block_t *block= chunk->contains_zip(data))
|
|
|
|
return block;
|
|
|
|
return nullptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Assert that all buffer pool pages are in a replaceable state */
|
|
|
|
void assert_all_freed();
|
|
|
|
#endif /* UNIV_DEBUG */
|
|
|
|
|
|
|
|
#ifdef BTR_CUR_HASH_ADAPT
|
|
|
|
/** Clear the adaptive hash index on all pages in the buffer pool. */
|
|
|
|
inline void clear_hash_index();
|
|
|
|
|
|
|
|
/** Get a buffer block from an adaptive hash index pointer.
|
|
|
|
This function does not return if the block is not identified.
|
|
|
|
@param ptr pointer to within a page frame
|
|
|
|
@return pointer to block, never NULL */
|
2020-06-12 12:50:31 +03:00
|
|
|
inline buf_block_t *block_from_ahi(const byte *ptr) const;
|
2020-03-18 21:48:00 +02:00
|
|
|
#endif /* BTR_CUR_HASH_ADAPT */
|
|
|
|
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
bool is_block_lock(const rw_lock_t *l) const
|
|
|
|
{ return is_block_field(static_cast<const void*>(l)); }
|
2020-03-18 21:48:00 +02:00
|
|
|
|
2020-06-02 08:17:10 +03:00
|
|
|
/**
|
|
|
|
@return the smallest oldest_modification lsn for any page
|
MDEV-23855: Improve InnoDB log checkpoint performance
After MDEV-15053, MDEV-22871, MDEV-23399 shifted the scalability
bottleneck, log checkpoints became a new bottleneck.
If innodb_io_capacity is set low or innodb_max_dirty_pct_lwm is
set high and the workload fits in the buffer pool, the page cleaner
thread will perform very little flushing. When we reach the capacity
of the circular redo log file ib_logfile0 and must initiate a checkpoint,
some 'furious flushing' will be necessary. (If innodb_flush_sync=OFF,
then flushing would continue at the innodb_io_capacity rate, and
writers would be throttled.)
We have the best chance of advancing the checkpoint LSN immediately
after a page flush batch has been completed. Hence, it is best to
perform checkpoints after every batch in the page cleaner thread,
attempting to run once per second.
By initiating high-priority flushing in the page cleaner as early
as possible, we aim to make the throughput more stable.
The function buf_flush_wait_flushed() used to sleep for 10ms, hoping
that the page cleaner thread would do something during that time.
The observed end result was that a large number of threads that call
log_free_check() would end up sleeping while nothing useful is happening.
We will revise the design so that in the default innodb_flush_sync=ON
mode, buf_flush_wait_flushed() will wake up the page cleaner thread
to perform the necessary flushing, and it will wait for a signal from
the page cleaner thread.
If innodb_io_capacity is set to a low value (causing the page cleaner to
throttle its work), a write workload would initially perform well, until
the capacity of the circular ib_logfile0 is reached and log_free_check()
will trigger checkpoints. At that point, the extra waiting in
buf_flush_wait_flushed() will start reducing throughput.
The page cleaner thread will also initiate log checkpoints after each
buf_flush_lists() call, because that is the best point of time for
the checkpoint LSN to advance by the maximum amount.
Even in 'furious flushing' mode we invoke buf_flush_lists() with
innodb_io_capacity_max pages at a time, and at the start of each
batch (in the log_flush() callback function that runs in a separate
task) we will invoke os_aio_wait_until_no_pending_writes(). This
tweak allows the checkpoint to advance in smaller steps and
significantly reduces the maximum latency. On an Intel Optane 960
NVMe SSD on Linux, it reduced from 4.6 seconds to 74 milliseconds.
On Microsoft Windows with a slower SSD, it reduced from more than
180 seconds to 0.6 seconds.
We will make innodb_adaptive_flushing=OFF simply flush innodb_io_capacity
per second whenever the dirty proportion of buffer pool pages exceeds
innodb_max_dirty_pages_pct_lwm. For innodb_adaptive_flushing=ON we try
to make page_cleaner_flush_pages_recommendation() more consistent and
predictable: if we are below innodb_adaptive_flushing_lwm, let us flush
pages according to the return value of af_get_pct_for_dirty().
innodb_max_dirty_pages_pct_lwm: Revert the change of the default value
that was made in MDEV-23399. The value innodb_max_dirty_pages_pct_lwm=0
guarantees that a shutdown of an idle server will be fast. Users might
be surprised if normal shutdown suddenly became slower when upgrading
within a GA release series.
innodb_checkpoint_usec: Remove. The master task will no longer perform
periodic log checkpoints. It is the duty of the page cleaner thread.
log_sys.max_modified_age: Remove. The current span of the
buf_pool.flush_list expressed in LSN only matters for adaptive
flushing (outside the 'furious flushing' condition).
For the correctness of checkpoints, the only thing that matters is
the checkpoint age (log_sys.lsn - log_sys.last_checkpoint_lsn).
This run-time constant was also reported as log_max_modified_age_sync.
log_sys.max_checkpoint_age_async: Remove. This does not serve any
purpose, because the checkpoints will now be triggered by the page
cleaner thread. We will retain the log_sys.max_checkpoint_age limit
for engaging 'furious flushing'.
page_cleaner.slot: Remove. It turns out that
page_cleaner_slot.flush_list_time was duplicating
page_cleaner.slot.flush_time and page_cleaner.slot.flush_list_pass
was duplicating page_cleaner.flush_pass.
Likewise, there were some redundant monitor counters, because the
page cleaner thread no longer performs any buf_pool.LRU flushing, and
because there only is one buf_flush_page_cleaner thread.
buf_flush_sync_lsn: Protect writes by buf_pool.flush_list_mutex.
buf_pool_t::get_oldest_modification(): Add a parameter to specify the
return value when no persistent data pages are dirty. Require the
caller to hold buf_pool.flush_list_mutex.
log_buf_pool_get_oldest_modification(): Take the fall-back LSN
as a parameter. All callers will also invoke log_sys.get_lsn().
log_preflush_pool_modified_pages(): Replaced with buf_flush_wait_flushed().
buf_flush_wait_flushed(): Implement two limits. If not enough buffer pool
has been flushed, signal the page cleaner (unless innodb_flush_sync=OFF)
and wait for the page cleaner to complete. If the page cleaner
thread is not running (which can be the case durign shutdown),
initiate the flush and wait for it directly.
buf_flush_ahead(): If innodb_flush_sync=ON (the default),
submit a new buf_flush_sync_lsn target for the page cleaner
but do not wait for the flushing to finish.
log_get_capacity(), log_get_max_modified_age_async(): Remove, to make
it easier to see that af_get_pct_for_lsn() is not acquiring any mutexes.
page_cleaner_flush_pages_recommendation(): Protect all access to
buf_pool.flush_list with buf_pool.flush_list_mutex. Previously there
were some race conditions in the calculation.
buf_flush_sync_for_checkpoint(): New function to process
buf_flush_sync_lsn in the page cleaner thread. At the end of
each batch, we try to wake up any blocked buf_flush_wait_flushed().
If everything up to buf_flush_sync_lsn has been flushed, we will
reset buf_flush_sync_lsn=0. The page cleaner thread will keep
'furious flushing' until the limit is reached. Any threads that
are waiting in buf_flush_wait_flushed() will be able to resume
as soon as their own limit has been satisfied.
buf_flush_page_cleaner: Prioritize buf_flush_sync_lsn and do not
sleep as long as it is set. Do not update any page_cleaner statistics
for this special mode of operation. In the normal mode
(buf_flush_sync_lsn is not set for innodb_flush_sync=ON),
try to wake up once per second. No longer check whether
srv_inc_activity_count() has been called. After each batch,
try to perform a log checkpoint, because the best chances for
the checkpoint LSN to advance by the maximum amount are upon
completing a flushing batch.
log_t: Move buf_free, max_buf_free possibly to the same cache line
with log_sys.mutex.
log_margin_checkpoint_age(): Simplify the logic, and replace
a 0.1-second sleep with a call to buf_flush_wait_flushed() to
initiate flushing. Moved to the same compilation unit
with the only caller.
log_close(): Clean up the calculations. (Should be no functional
change.) Return whether flush-ahead is needed. Moved to the same
compilation unit with the only caller.
mtr_t::finish_write(): Return whether flush-ahead is needed.
mtr_t::commit(): Invoke buf_flush_ahead() when needed. Let us avoid
external calls in mtr_t::commit() and make the logic easier to follow
by having related code in a single compilation unit. Also, we will
invoke srv_stats.log_write_requests.inc() only once per
mini-transaction commit, while not holding mutexes.
log_checkpoint_margin(): Only care about log_sys.max_checkpoint_age.
Upon reaching log_sys.max_checkpoint_age where we must wait to prevent
the log from getting corrupted, let us wait for at most 1MiB of LSN
at a time, before rechecking the condition. This should allow writers
to proceed even if the redo log capacity has been reached and
'furious flushing' is in progress. We no longer care about
log_sys.max_modified_age_sync or log_sys.max_modified_age_async.
The log_sys.max_modified_age_sync could be a relic from the time when
there was a srv_master_thread that wrote dirty pages to data files.
Also, we no longer have any log_sys.max_checkpoint_age_async limit,
because log checkpoints will now be triggered by the page cleaner
thread upon completing buf_flush_lists().
log_set_capacity(): Simplify the calculations of the limit
(no functional change).
log_checkpoint_low(): Split from log_checkpoint(). Moved to the
same compilation unit with the caller.
log_make_checkpoint(): Only wait for everything to be flushed until
the current LSN.
create_log_file(): After checkpoint, invoke log_write_up_to()
to ensure that the FILE_CHECKPOINT record has been written.
This avoids ut_ad(!srv_log_file_created) in create_log_file_rename().
srv_start(): Do not call recv_recovery_from_checkpoint_start()
if the log has just been created. Set fil_system.space_id_reuse_warned
before dict_boot() has been executed, and clear it after recovery
has finished.
dict_boot(): Initialize fil_system.max_assigned_id.
srv_check_activity(): Remove. The activity count is counting transaction
commits and therefore mostly interesting for the purge of history.
BtrBulk::insert(): Do not explicitly wake up the page cleaner,
but do invoke srv_inc_activity_count(), because that counter is
still being used in buf_load_throttle_if_needed() for some
heuristics. (It might be cleaner to execute buf_load() in the
page cleaner thread!)
Reviewed by: Vladislav Vaintroub
2020-10-26 16:35:47 +02:00
|
|
|
@retval empty_lsn if all modified persistent pages have been flushed */
|
2021-06-23 13:13:11 +03:00
|
|
|
lsn_t get_oldest_modification(lsn_t empty_lsn)
|
MDEV-23855: Improve InnoDB log checkpoint performance
After MDEV-15053, MDEV-22871, MDEV-23399 shifted the scalability
bottleneck, log checkpoints became a new bottleneck.
If innodb_io_capacity is set low or innodb_max_dirty_pct_lwm is
set high and the workload fits in the buffer pool, the page cleaner
thread will perform very little flushing. When we reach the capacity
of the circular redo log file ib_logfile0 and must initiate a checkpoint,
some 'furious flushing' will be necessary. (If innodb_flush_sync=OFF,
then flushing would continue at the innodb_io_capacity rate, and
writers would be throttled.)
We have the best chance of advancing the checkpoint LSN immediately
after a page flush batch has been completed. Hence, it is best to
perform checkpoints after every batch in the page cleaner thread,
attempting to run once per second.
By initiating high-priority flushing in the page cleaner as early
as possible, we aim to make the throughput more stable.
The function buf_flush_wait_flushed() used to sleep for 10ms, hoping
that the page cleaner thread would do something during that time.
The observed end result was that a large number of threads that call
log_free_check() would end up sleeping while nothing useful is happening.
We will revise the design so that in the default innodb_flush_sync=ON
mode, buf_flush_wait_flushed() will wake up the page cleaner thread
to perform the necessary flushing, and it will wait for a signal from
the page cleaner thread.
If innodb_io_capacity is set to a low value (causing the page cleaner to
throttle its work), a write workload would initially perform well, until
the capacity of the circular ib_logfile0 is reached and log_free_check()
will trigger checkpoints. At that point, the extra waiting in
buf_flush_wait_flushed() will start reducing throughput.
The page cleaner thread will also initiate log checkpoints after each
buf_flush_lists() call, because that is the best point of time for
the checkpoint LSN to advance by the maximum amount.
Even in 'furious flushing' mode we invoke buf_flush_lists() with
innodb_io_capacity_max pages at a time, and at the start of each
batch (in the log_flush() callback function that runs in a separate
task) we will invoke os_aio_wait_until_no_pending_writes(). This
tweak allows the checkpoint to advance in smaller steps and
significantly reduces the maximum latency. On an Intel Optane 960
NVMe SSD on Linux, it reduced from 4.6 seconds to 74 milliseconds.
On Microsoft Windows with a slower SSD, it reduced from more than
180 seconds to 0.6 seconds.
We will make innodb_adaptive_flushing=OFF simply flush innodb_io_capacity
per second whenever the dirty proportion of buffer pool pages exceeds
innodb_max_dirty_pages_pct_lwm. For innodb_adaptive_flushing=ON we try
to make page_cleaner_flush_pages_recommendation() more consistent and
predictable: if we are below innodb_adaptive_flushing_lwm, let us flush
pages according to the return value of af_get_pct_for_dirty().
innodb_max_dirty_pages_pct_lwm: Revert the change of the default value
that was made in MDEV-23399. The value innodb_max_dirty_pages_pct_lwm=0
guarantees that a shutdown of an idle server will be fast. Users might
be surprised if normal shutdown suddenly became slower when upgrading
within a GA release series.
innodb_checkpoint_usec: Remove. The master task will no longer perform
periodic log checkpoints. It is the duty of the page cleaner thread.
log_sys.max_modified_age: Remove. The current span of the
buf_pool.flush_list expressed in LSN only matters for adaptive
flushing (outside the 'furious flushing' condition).
For the correctness of checkpoints, the only thing that matters is
the checkpoint age (log_sys.lsn - log_sys.last_checkpoint_lsn).
This run-time constant was also reported as log_max_modified_age_sync.
log_sys.max_checkpoint_age_async: Remove. This does not serve any
purpose, because the checkpoints will now be triggered by the page
cleaner thread. We will retain the log_sys.max_checkpoint_age limit
for engaging 'furious flushing'.
page_cleaner.slot: Remove. It turns out that
page_cleaner_slot.flush_list_time was duplicating
page_cleaner.slot.flush_time and page_cleaner.slot.flush_list_pass
was duplicating page_cleaner.flush_pass.
Likewise, there were some redundant monitor counters, because the
page cleaner thread no longer performs any buf_pool.LRU flushing, and
because there only is one buf_flush_page_cleaner thread.
buf_flush_sync_lsn: Protect writes by buf_pool.flush_list_mutex.
buf_pool_t::get_oldest_modification(): Add a parameter to specify the
return value when no persistent data pages are dirty. Require the
caller to hold buf_pool.flush_list_mutex.
log_buf_pool_get_oldest_modification(): Take the fall-back LSN
as a parameter. All callers will also invoke log_sys.get_lsn().
log_preflush_pool_modified_pages(): Replaced with buf_flush_wait_flushed().
buf_flush_wait_flushed(): Implement two limits. If not enough buffer pool
has been flushed, signal the page cleaner (unless innodb_flush_sync=OFF)
and wait for the page cleaner to complete. If the page cleaner
thread is not running (which can be the case durign shutdown),
initiate the flush and wait for it directly.
buf_flush_ahead(): If innodb_flush_sync=ON (the default),
submit a new buf_flush_sync_lsn target for the page cleaner
but do not wait for the flushing to finish.
log_get_capacity(), log_get_max_modified_age_async(): Remove, to make
it easier to see that af_get_pct_for_lsn() is not acquiring any mutexes.
page_cleaner_flush_pages_recommendation(): Protect all access to
buf_pool.flush_list with buf_pool.flush_list_mutex. Previously there
were some race conditions in the calculation.
buf_flush_sync_for_checkpoint(): New function to process
buf_flush_sync_lsn in the page cleaner thread. At the end of
each batch, we try to wake up any blocked buf_flush_wait_flushed().
If everything up to buf_flush_sync_lsn has been flushed, we will
reset buf_flush_sync_lsn=0. The page cleaner thread will keep
'furious flushing' until the limit is reached. Any threads that
are waiting in buf_flush_wait_flushed() will be able to resume
as soon as their own limit has been satisfied.
buf_flush_page_cleaner: Prioritize buf_flush_sync_lsn and do not
sleep as long as it is set. Do not update any page_cleaner statistics
for this special mode of operation. In the normal mode
(buf_flush_sync_lsn is not set for innodb_flush_sync=ON),
try to wake up once per second. No longer check whether
srv_inc_activity_count() has been called. After each batch,
try to perform a log checkpoint, because the best chances for
the checkpoint LSN to advance by the maximum amount are upon
completing a flushing batch.
log_t: Move buf_free, max_buf_free possibly to the same cache line
with log_sys.mutex.
log_margin_checkpoint_age(): Simplify the logic, and replace
a 0.1-second sleep with a call to buf_flush_wait_flushed() to
initiate flushing. Moved to the same compilation unit
with the only caller.
log_close(): Clean up the calculations. (Should be no functional
change.) Return whether flush-ahead is needed. Moved to the same
compilation unit with the only caller.
mtr_t::finish_write(): Return whether flush-ahead is needed.
mtr_t::commit(): Invoke buf_flush_ahead() when needed. Let us avoid
external calls in mtr_t::commit() and make the logic easier to follow
by having related code in a single compilation unit. Also, we will
invoke srv_stats.log_write_requests.inc() only once per
mini-transaction commit, while not holding mutexes.
log_checkpoint_margin(): Only care about log_sys.max_checkpoint_age.
Upon reaching log_sys.max_checkpoint_age where we must wait to prevent
the log from getting corrupted, let us wait for at most 1MiB of LSN
at a time, before rechecking the condition. This should allow writers
to proceed even if the redo log capacity has been reached and
'furious flushing' is in progress. We no longer care about
log_sys.max_modified_age_sync or log_sys.max_modified_age_async.
The log_sys.max_modified_age_sync could be a relic from the time when
there was a srv_master_thread that wrote dirty pages to data files.
Also, we no longer have any log_sys.max_checkpoint_age_async limit,
because log checkpoints will now be triggered by the page cleaner
thread upon completing buf_flush_lists().
log_set_capacity(): Simplify the calculations of the limit
(no functional change).
log_checkpoint_low(): Split from log_checkpoint(). Moved to the
same compilation unit with the caller.
log_make_checkpoint(): Only wait for everything to be flushed until
the current LSN.
create_log_file(): After checkpoint, invoke log_write_up_to()
to ensure that the FILE_CHECKPOINT record has been written.
This avoids ut_ad(!srv_log_file_created) in create_log_file_rename().
srv_start(): Do not call recv_recovery_from_checkpoint_start()
if the log has just been created. Set fil_system.space_id_reuse_warned
before dict_boot() has been executed, and clear it after recovery
has finished.
dict_boot(): Initialize fil_system.max_assigned_id.
srv_check_activity(): Remove. The activity count is counting transaction
commits and therefore mostly interesting for the purge of history.
BtrBulk::insert(): Do not explicitly wake up the page cleaner,
but do invoke srv_inc_activity_count(), because that counter is
still being used in buf_load_throttle_if_needed() for some
heuristics. (It might be cleaner to execute buf_load() in the
page cleaner thread!)
Reviewed by: Vladislav Vaintroub
2020-10-26 16:35:47 +02:00
|
|
|
{
|
2021-06-23 13:13:11 +03:00
|
|
|
mysql_mutex_assert_owner(&flush_list_mutex);
|
|
|
|
while (buf_page_t *bpage= UT_LIST_GET_LAST(flush_list))
|
|
|
|
{
|
|
|
|
ut_ad(!fsp_is_system_temporary(bpage->id().space()));
|
|
|
|
lsn_t lsn= bpage->oldest_modification();
|
|
|
|
if (lsn != 1)
|
|
|
|
{
|
|
|
|
ut_ad(lsn > 2);
|
|
|
|
return lsn;
|
|
|
|
}
|
|
|
|
delete_from_flush_list(bpage);
|
|
|
|
}
|
|
|
|
return empty_lsn;
|
MDEV-23855: Improve InnoDB log checkpoint performance
After MDEV-15053, MDEV-22871, MDEV-23399 shifted the scalability
bottleneck, log checkpoints became a new bottleneck.
If innodb_io_capacity is set low or innodb_max_dirty_pct_lwm is
set high and the workload fits in the buffer pool, the page cleaner
thread will perform very little flushing. When we reach the capacity
of the circular redo log file ib_logfile0 and must initiate a checkpoint,
some 'furious flushing' will be necessary. (If innodb_flush_sync=OFF,
then flushing would continue at the innodb_io_capacity rate, and
writers would be throttled.)
We have the best chance of advancing the checkpoint LSN immediately
after a page flush batch has been completed. Hence, it is best to
perform checkpoints after every batch in the page cleaner thread,
attempting to run once per second.
By initiating high-priority flushing in the page cleaner as early
as possible, we aim to make the throughput more stable.
The function buf_flush_wait_flushed() used to sleep for 10ms, hoping
that the page cleaner thread would do something during that time.
The observed end result was that a large number of threads that call
log_free_check() would end up sleeping while nothing useful is happening.
We will revise the design so that in the default innodb_flush_sync=ON
mode, buf_flush_wait_flushed() will wake up the page cleaner thread
to perform the necessary flushing, and it will wait for a signal from
the page cleaner thread.
If innodb_io_capacity is set to a low value (causing the page cleaner to
throttle its work), a write workload would initially perform well, until
the capacity of the circular ib_logfile0 is reached and log_free_check()
will trigger checkpoints. At that point, the extra waiting in
buf_flush_wait_flushed() will start reducing throughput.
The page cleaner thread will also initiate log checkpoints after each
buf_flush_lists() call, because that is the best point of time for
the checkpoint LSN to advance by the maximum amount.
Even in 'furious flushing' mode we invoke buf_flush_lists() with
innodb_io_capacity_max pages at a time, and at the start of each
batch (in the log_flush() callback function that runs in a separate
task) we will invoke os_aio_wait_until_no_pending_writes(). This
tweak allows the checkpoint to advance in smaller steps and
significantly reduces the maximum latency. On an Intel Optane 960
NVMe SSD on Linux, it reduced from 4.6 seconds to 74 milliseconds.
On Microsoft Windows with a slower SSD, it reduced from more than
180 seconds to 0.6 seconds.
We will make innodb_adaptive_flushing=OFF simply flush innodb_io_capacity
per second whenever the dirty proportion of buffer pool pages exceeds
innodb_max_dirty_pages_pct_lwm. For innodb_adaptive_flushing=ON we try
to make page_cleaner_flush_pages_recommendation() more consistent and
predictable: if we are below innodb_adaptive_flushing_lwm, let us flush
pages according to the return value of af_get_pct_for_dirty().
innodb_max_dirty_pages_pct_lwm: Revert the change of the default value
that was made in MDEV-23399. The value innodb_max_dirty_pages_pct_lwm=0
guarantees that a shutdown of an idle server will be fast. Users might
be surprised if normal shutdown suddenly became slower when upgrading
within a GA release series.
innodb_checkpoint_usec: Remove. The master task will no longer perform
periodic log checkpoints. It is the duty of the page cleaner thread.
log_sys.max_modified_age: Remove. The current span of the
buf_pool.flush_list expressed in LSN only matters for adaptive
flushing (outside the 'furious flushing' condition).
For the correctness of checkpoints, the only thing that matters is
the checkpoint age (log_sys.lsn - log_sys.last_checkpoint_lsn).
This run-time constant was also reported as log_max_modified_age_sync.
log_sys.max_checkpoint_age_async: Remove. This does not serve any
purpose, because the checkpoints will now be triggered by the page
cleaner thread. We will retain the log_sys.max_checkpoint_age limit
for engaging 'furious flushing'.
page_cleaner.slot: Remove. It turns out that
page_cleaner_slot.flush_list_time was duplicating
page_cleaner.slot.flush_time and page_cleaner.slot.flush_list_pass
was duplicating page_cleaner.flush_pass.
Likewise, there were some redundant monitor counters, because the
page cleaner thread no longer performs any buf_pool.LRU flushing, and
because there only is one buf_flush_page_cleaner thread.
buf_flush_sync_lsn: Protect writes by buf_pool.flush_list_mutex.
buf_pool_t::get_oldest_modification(): Add a parameter to specify the
return value when no persistent data pages are dirty. Require the
caller to hold buf_pool.flush_list_mutex.
log_buf_pool_get_oldest_modification(): Take the fall-back LSN
as a parameter. All callers will also invoke log_sys.get_lsn().
log_preflush_pool_modified_pages(): Replaced with buf_flush_wait_flushed().
buf_flush_wait_flushed(): Implement two limits. If not enough buffer pool
has been flushed, signal the page cleaner (unless innodb_flush_sync=OFF)
and wait for the page cleaner to complete. If the page cleaner
thread is not running (which can be the case durign shutdown),
initiate the flush and wait for it directly.
buf_flush_ahead(): If innodb_flush_sync=ON (the default),
submit a new buf_flush_sync_lsn target for the page cleaner
but do not wait for the flushing to finish.
log_get_capacity(), log_get_max_modified_age_async(): Remove, to make
it easier to see that af_get_pct_for_lsn() is not acquiring any mutexes.
page_cleaner_flush_pages_recommendation(): Protect all access to
buf_pool.flush_list with buf_pool.flush_list_mutex. Previously there
were some race conditions in the calculation.
buf_flush_sync_for_checkpoint(): New function to process
buf_flush_sync_lsn in the page cleaner thread. At the end of
each batch, we try to wake up any blocked buf_flush_wait_flushed().
If everything up to buf_flush_sync_lsn has been flushed, we will
reset buf_flush_sync_lsn=0. The page cleaner thread will keep
'furious flushing' until the limit is reached. Any threads that
are waiting in buf_flush_wait_flushed() will be able to resume
as soon as their own limit has been satisfied.
buf_flush_page_cleaner: Prioritize buf_flush_sync_lsn and do not
sleep as long as it is set. Do not update any page_cleaner statistics
for this special mode of operation. In the normal mode
(buf_flush_sync_lsn is not set for innodb_flush_sync=ON),
try to wake up once per second. No longer check whether
srv_inc_activity_count() has been called. After each batch,
try to perform a log checkpoint, because the best chances for
the checkpoint LSN to advance by the maximum amount are upon
completing a flushing batch.
log_t: Move buf_free, max_buf_free possibly to the same cache line
with log_sys.mutex.
log_margin_checkpoint_age(): Simplify the logic, and replace
a 0.1-second sleep with a call to buf_flush_wait_flushed() to
initiate flushing. Moved to the same compilation unit
with the only caller.
log_close(): Clean up the calculations. (Should be no functional
change.) Return whether flush-ahead is needed. Moved to the same
compilation unit with the only caller.
mtr_t::finish_write(): Return whether flush-ahead is needed.
mtr_t::commit(): Invoke buf_flush_ahead() when needed. Let us avoid
external calls in mtr_t::commit() and make the logic easier to follow
by having related code in a single compilation unit. Also, we will
invoke srv_stats.log_write_requests.inc() only once per
mini-transaction commit, while not holding mutexes.
log_checkpoint_margin(): Only care about log_sys.max_checkpoint_age.
Upon reaching log_sys.max_checkpoint_age where we must wait to prevent
the log from getting corrupted, let us wait for at most 1MiB of LSN
at a time, before rechecking the condition. This should allow writers
to proceed even if the redo log capacity has been reached and
'furious flushing' is in progress. We no longer care about
log_sys.max_modified_age_sync or log_sys.max_modified_age_async.
The log_sys.max_modified_age_sync could be a relic from the time when
there was a srv_master_thread that wrote dirty pages to data files.
Also, we no longer have any log_sys.max_checkpoint_age_async limit,
because log checkpoints will now be triggered by the page cleaner
thread upon completing buf_flush_lists().
log_set_capacity(): Simplify the calculations of the limit
(no functional change).
log_checkpoint_low(): Split from log_checkpoint(). Moved to the
same compilation unit with the caller.
log_make_checkpoint(): Only wait for everything to be flushed until
the current LSN.
create_log_file(): After checkpoint, invoke log_write_up_to()
to ensure that the FILE_CHECKPOINT record has been written.
This avoids ut_ad(!srv_log_file_created) in create_log_file_rename().
srv_start(): Do not call recv_recovery_from_checkpoint_start()
if the log has just been created. Set fil_system.space_id_reuse_warned
before dict_boot() has been executed, and clear it after recovery
has finished.
dict_boot(): Initialize fil_system.max_assigned_id.
srv_check_activity(): Remove. The activity count is counting transaction
commits and therefore mostly interesting for the purge of history.
BtrBulk::insert(): Do not explicitly wake up the page cleaner,
but do invoke srv_inc_activity_count(), because that counter is
still being used in buf_load_throttle_if_needed() for some
heuristics. (It might be cleaner to execute buf_load() in the
page cleaner thread!)
Reviewed by: Vladislav Vaintroub
2020-10-26 16:35:47 +02:00
|
|
|
}
|
2020-06-02 08:17:10 +03:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Determine if a buffer block was created by chunk_t::create().
|
|
|
|
@param block block descriptor (not dereferenced)
|
|
|
|
@return whether block has been created by chunk_t::create() */
|
|
|
|
bool is_uncompressed(const buf_block_t *block) const
|
|
|
|
{
|
2020-10-30 11:15:30 +02:00
|
|
|
return is_block_field(reinterpret_cast<const void*>(block));
|
2020-03-18 21:48:00 +02:00
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Get the page_hash latch for a page */
|
2020-06-18 13:38:30 +03:00
|
|
|
page_hash_latch *hash_lock_get(const page_id_t id) const
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
{
|
2020-06-18 13:38:30 +03:00
|
|
|
return page_hash.lock_get(id.fold());
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
/** Look up a block descriptor.
|
2020-06-12 12:50:31 +03:00
|
|
|
@param id page identifier
|
|
|
|
@param fold id.fold()
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
@return block descriptor, possibly in watch[]
|
|
|
|
@retval nullptr if not found*/
|
2020-06-12 12:50:31 +03:00
|
|
|
buf_page_t *page_hash_get_low(const page_id_t id, const ulint fold)
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
{
|
2020-06-12 12:50:31 +03:00
|
|
|
ut_ad(id.fold() == fold);
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
#ifdef SAFE_MUTEX
|
|
|
|
DBUG_ASSERT(mysql_mutex_is_owner(&mutex) ||
|
|
|
|
page_hash.lock_get(fold)->is_locked());
|
|
|
|
#endif /* SAFE_MUTEX */
|
2020-06-12 12:50:31 +03:00
|
|
|
buf_page_t *bpage;
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/* Look for the page in the hash table */
|
2020-06-18 12:26:28 +03:00
|
|
|
HASH_SEARCH(hash, &page_hash, fold, buf_page_t*, bpage,
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(bpage->in_page_hash), id == bpage->id());
|
|
|
|
return bpage;
|
|
|
|
}
|
2020-06-12 12:50:31 +03:00
|
|
|
private:
|
|
|
|
/** Look up a block descriptor.
|
|
|
|
@tparam exclusive whether the latch is to be acquired exclusively
|
|
|
|
@tparam watch whether to allow watch_is_sentinel()
|
|
|
|
@param page_id page identifier
|
|
|
|
@param fold page_id.fold()
|
|
|
|
@param hash_lock pointer to the acquired latch (to be released by caller)
|
|
|
|
@return pointer to the block
|
|
|
|
@retval nullptr if no block was found; !lock || !*lock will also hold */
|
|
|
|
template<bool exclusive,bool watch>
|
|
|
|
buf_page_t *page_hash_get_locked(const page_id_t page_id, ulint fold,
|
2020-06-18 13:38:30 +03:00
|
|
|
page_hash_latch **hash_lock)
|
2020-06-12 12:50:31 +03:00
|
|
|
{
|
|
|
|
ut_ad(hash_lock || !exclusive);
|
2020-06-18 13:38:30 +03:00
|
|
|
page_hash_latch *latch= page_hash.lock<exclusive>(fold);
|
2020-06-12 12:50:31 +03:00
|
|
|
buf_page_t *bpage= page_hash_get_low(page_id, fold);
|
|
|
|
if (!bpage || watch_is_sentinel(*bpage))
|
|
|
|
{
|
2020-06-18 13:38:30 +03:00
|
|
|
latch->release<exclusive>();
|
2020-06-12 12:50:31 +03:00
|
|
|
if (hash_lock)
|
|
|
|
*hash_lock= nullptr;
|
|
|
|
return watch ? bpage : nullptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
ut_ad(bpage->in_file());
|
|
|
|
ut_ad(page_id == bpage->id());
|
|
|
|
|
|
|
|
if (hash_lock)
|
|
|
|
*hash_lock= latch; /* to be released by the caller */
|
|
|
|
else
|
2020-06-18 13:38:30 +03:00
|
|
|
latch->release<exclusive>();
|
2020-06-12 12:50:31 +03:00
|
|
|
return bpage;
|
|
|
|
}
|
|
|
|
public:
|
|
|
|
/** Look up a block descriptor.
|
|
|
|
@tparam exclusive whether the latch is to be acquired exclusively
|
|
|
|
@param page_id page identifier
|
|
|
|
@param fold page_id.fold()
|
|
|
|
@param hash_lock pointer to the acquired latch (to be released by caller)
|
|
|
|
@return pointer to the block
|
|
|
|
@retval nullptr if no block was found; !lock || !*lock will also hold */
|
|
|
|
template<bool exclusive>
|
|
|
|
buf_page_t *page_hash_get_locked(const page_id_t page_id, ulint fold,
|
2020-06-18 13:38:30 +03:00
|
|
|
page_hash_latch **hash_lock)
|
2020-06-12 12:50:31 +03:00
|
|
|
{ return page_hash_get_locked<exclusive,false>(page_id, fold, hash_lock); }
|
|
|
|
|
|
|
|
/** @return whether the buffer pool contains a page
|
|
|
|
@tparam watch whether to allow watch_is_sentinel()
|
|
|
|
@param page_id page identifier */
|
|
|
|
template<bool watch= false>
|
|
|
|
bool page_hash_contains(const page_id_t page_id)
|
|
|
|
{
|
|
|
|
return page_hash_get_locked<false,watch>(page_id, page_id.fold(), nullptr);
|
|
|
|
}
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
|
|
|
|
/** Determine if a block is a sentinel for a buffer pool watch.
|
|
|
|
@param bpage page descriptor
|
|
|
|
@return whether bpage a sentinel for a buffer pool watch */
|
|
|
|
bool watch_is_sentinel(const buf_page_t &bpage)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
#ifdef SAFE_MUTEX
|
|
|
|
DBUG_ASSERT(mysql_mutex_is_owner(&mutex) ||
|
|
|
|
hash_lock_get(bpage.id())->is_locked());
|
|
|
|
#endif /* SAFE_MUTEX */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(bpage.in_file());
|
|
|
|
|
|
|
|
if (&bpage < &watch[0] || &bpage >= &watch[UT_ARR_SIZE(watch)])
|
|
|
|
{
|
|
|
|
ut_ad(bpage.state() != BUF_BLOCK_ZIP_PAGE || bpage.zip.data);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
ut_ad(bpage.state() == BUF_BLOCK_ZIP_PAGE);
|
|
|
|
ut_ad(!bpage.in_zip_hash);
|
|
|
|
ut_ad(!bpage.zip.data);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Check if a watched page has been read.
|
|
|
|
This may only be called after !watch_set() and before invoking watch_unset().
|
|
|
|
@param id page identifier
|
|
|
|
@return whether the page was read to the buffer pool */
|
|
|
|
bool watch_occurred(const page_id_t id)
|
|
|
|
{
|
2020-06-12 12:50:31 +03:00
|
|
|
const ulint fold= id.fold();
|
2020-06-18 13:38:30 +03:00
|
|
|
page_hash_latch *hash_lock= page_hash.lock<false>(fold);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/* The page must exist because watch_set() increments buf_fix_count. */
|
2020-06-12 12:50:31 +03:00
|
|
|
buf_page_t *bpage= page_hash_get_low(id, fold);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
const bool is_sentinel= watch_is_sentinel(*bpage);
|
2020-06-18 13:38:30 +03:00
|
|
|
hash_lock->read_unlock();
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
return !is_sentinel;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Register a watch for a page identifier. The caller must hold an
|
|
|
|
exclusive page hash latch. The *hash_lock may be released,
|
|
|
|
relocated, and reacquired.
|
|
|
|
@param id page identifier
|
2020-06-12 12:50:31 +03:00
|
|
|
@param hash_lock exclusively held page_hash latch
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
@return a buffer pool block corresponding to id
|
|
|
|
@retval nullptr if the block was not present, and a watch was installed */
|
2020-06-18 13:38:30 +03:00
|
|
|
inline buf_page_t *watch_set(const page_id_t id,
|
|
|
|
page_hash_latch **hash_lock);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
|
|
|
|
/** Stop watching whether a page has been read in.
|
|
|
|
watch_set(id) must have returned nullptr before.
|
|
|
|
@param id page identifier */
|
2021-09-11 11:45:17 +03:00
|
|
|
void watch_unset(const page_id_t id);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
|
|
|
|
/** Remove the sentinel block for the watch before replacing it with a
|
|
|
|
real block. watch_unset() or watch_occurred() will notice
|
|
|
|
that the block has been replaced with the real block.
|
|
|
|
@param watch sentinel */
|
|
|
|
inline void watch_remove(buf_page_t *watch);
|
|
|
|
|
|
|
|
/** @return whether less than 1/4 of the buffer pool is available */
|
|
|
|
bool running_out() const
|
|
|
|
{
|
|
|
|
return !recv_recovery_is_on() &&
|
|
|
|
UNIV_UNLIKELY(UT_LIST_GET_LEN(free) + UT_LIST_GET_LEN(LRU) <
|
|
|
|
std::min(curr_size, old_size) / 4);
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef UNIV_DEBUG
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Validate the buffer pool. */
|
|
|
|
void validate();
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
#endif /* UNIV_DEBUG */
|
|
|
|
#if defined UNIV_DEBUG_PRINT || defined UNIV_DEBUG
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Write information of the buf_pool to the error log. */
|
|
|
|
void print();
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
#endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG */
|
|
|
|
|
|
|
|
/** Remove a block from the LRU list.
|
|
|
|
@return the predecessor in the LRU list */
|
|
|
|
buf_page_t *LRU_remove(buf_page_t *bpage)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(bpage->in_LRU_list);
|
|
|
|
ut_ad(bpage->in_page_hash);
|
|
|
|
ut_ad(!bpage->in_zip_hash);
|
|
|
|
ut_ad(bpage->in_file());
|
|
|
|
lru_hp.adjust(bpage);
|
|
|
|
lru_scan_itr.adjust(bpage);
|
|
|
|
ut_d(bpage->in_LRU_list= false);
|
|
|
|
buf_page_t *prev= UT_LIST_GET_PREV(LRU, bpage);
|
|
|
|
UT_LIST_REMOVE(LRU, bpage);
|
|
|
|
return prev;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Number of pages to read ahead */
|
|
|
|
static constexpr uint32_t READ_AHEAD_PAGES= 64;
|
2020-03-18 21:48:00 +02:00
|
|
|
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
/** Buffer pool mutex */
|
2021-01-07 11:18:13 +02:00
|
|
|
MY_ALIGNED(CPU_LEVEL1_DCACHE_LINESIZE) mysql_mutex_t mutex;
|
2021-06-23 13:13:11 +03:00
|
|
|
/** Number of pending LRU flush; protected by mutex. */
|
|
|
|
ulint n_flush_LRU_;
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
/** broadcast when n_flush_LRU reaches 0; protected by mutex */
|
2021-02-07 12:19:24 +02:00
|
|
|
pthread_cond_t done_flush_LRU;
|
2021-06-23 13:13:11 +03:00
|
|
|
/** Number of pending flush_list flush; protected by mutex */
|
|
|
|
ulint n_flush_list_;
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
/** broadcast when n_flush_list reaches 0; protected by mutex */
|
2021-02-07 12:19:24 +02:00
|
|
|
pthread_cond_t done_flush_list;
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
|
2021-06-23 13:13:11 +03:00
|
|
|
TPOOL_SUPPRESS_TSAN ulint n_flush_LRU() const { return n_flush_LRU_; }
|
|
|
|
TPOOL_SUPPRESS_TSAN ulint n_flush_list() const { return n_flush_list_; }
|
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
/** @name General fields */
|
|
|
|
/* @{ */
|
|
|
|
ulint curr_pool_size; /*!< Current pool size in bytes */
|
|
|
|
ulint LRU_old_ratio; /*!< Reserve this much of the buffer
|
|
|
|
pool for "old" blocks */
|
|
|
|
#ifdef UNIV_DEBUG
|
|
|
|
ulint buddy_n_frames; /*!< Number of frames allocated from
|
|
|
|
the buffer pool to the buddy system */
|
|
|
|
ulint mutex_exit_forbidden; /*!< Forbid release mutex */
|
|
|
|
#endif
|
2016-08-12 11:17:45 +03:00
|
|
|
ut_allocator<unsigned char> allocator; /*!< Allocator used for
|
|
|
|
allocating memory for the the "chunks"
|
|
|
|
member. */
|
|
|
|
volatile ulint n_chunks; /*!< number of buffer pool chunks */
|
|
|
|
volatile ulint n_chunks_new; /*!< new number of buffer pool chunks */
|
2020-03-18 21:48:00 +02:00
|
|
|
chunk_t* chunks; /*!< buffer pool chunks */
|
|
|
|
chunk_t* chunks_old; /*!< old buffer pool chunks to be freed
|
2016-08-12 11:17:45 +03:00
|
|
|
after resizing buffer pool */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** current pool size in pages */
|
|
|
|
Atomic_counter<ulint> curr_size;
|
|
|
|
/** previous pool size in pages */
|
|
|
|
Atomic_counter<ulint> old_size;
|
|
|
|
/** read-ahead request size in pages */
|
|
|
|
Atomic_counter<uint32_t> read_ahead_area;
|
|
|
|
|
2020-06-18 13:38:30 +03:00
|
|
|
/** Hash table with singly-linked overflow lists. @see hash_table_t */
|
|
|
|
struct page_hash_table
|
|
|
|
{
|
|
|
|
/** Number of array[] elements per page_hash_latch.
|
|
|
|
Must be one less than a power of 2. */
|
MDEV-23369 False sharing in page_hash_latch::read_lock_wait()
MDEV-22871 refactored the InnoDB buf_pool.page_hash to use a simple
rw-lock implementation that avoids a spinloop between non-contended
read-lock requests, simply using std::atomic::fetch_add() for the
lock acquisition.
Alas, in a write-heavy stress test on a 56-core system with 1,000
concurrent client connections, the server would stop processing
any transactions every now and then. The reason turned out to be
false sharing. Attaching a debugger to the server during one such
hang revealed that 22 of the 1,033 threads were polling in
page_hash_latch::read_lock_wait() on the same object, which appeared
to be in unlocked state (no readers or writers). All 22 requests were
for accessing an undo log page, with a distinct page number.
To eliminate such false sharing, we will make buf_pool.page_hash.array
contain one page_hash_latch per CPU data cache line. On AMD64, this
will pad the size of the array by 8/7, or almost 15%. For a 50GiB
buffer pool of 16KiB pages, the buf_pool.page_hash.array would
grow from 25MiB to 28.6MiB. On other instruction set architectures,
the incurred memory overhead may be smaller.
Thanks to Vladislav Vaintroub for noticing this anomaly.
2020-08-02 20:39:36 +03:00
|
|
|
static constexpr size_t ELEMENTS_PER_LATCH= CPU_LEVEL1_DCACHE_LINESIZE /
|
|
|
|
sizeof(void*) - 1;
|
2020-06-18 13:38:30 +03:00
|
|
|
|
|
|
|
/** number of payload elements in array[] */
|
|
|
|
Atomic_relaxed<ulint> n_cells;
|
2020-07-13 17:53:12 +03:00
|
|
|
/** the hash table, with pad(n_cells) elements, aligned to L1 cache size */
|
2020-06-18 13:38:30 +03:00
|
|
|
hash_cell_t *array;
|
|
|
|
|
|
|
|
/** Create the hash table.
|
|
|
|
@param n the lower bound of n_cells */
|
|
|
|
void create(ulint n);
|
|
|
|
|
|
|
|
/** Free the hash table. */
|
2020-07-13 17:53:12 +03:00
|
|
|
void free() { aligned_free(array); array= nullptr; }
|
2020-06-18 13:38:30 +03:00
|
|
|
|
|
|
|
/** @return the index of an array element */
|
|
|
|
ulint calc_hash(ulint fold) const { return calc_hash(fold, n_cells); }
|
|
|
|
/** @return raw array index converted to padded index */
|
|
|
|
static ulint pad(ulint h) { return 1 + (h / ELEMENTS_PER_LATCH) + h; }
|
|
|
|
private:
|
|
|
|
/** @return the hash value before any ELEMENTS_PER_LATCH padding */
|
|
|
|
static ulint hash(ulint fold, ulint n) { return ut_hash_ulint(fold, n); }
|
|
|
|
|
|
|
|
/** @return the index of an array element */
|
|
|
|
static ulint calc_hash(ulint fold, ulint n_cells)
|
|
|
|
{
|
|
|
|
return pad(hash(fold, n_cells));
|
|
|
|
}
|
|
|
|
/** Get a page_hash latch. */
|
|
|
|
page_hash_latch *lock_get(ulint fold, ulint n) const
|
|
|
|
{
|
|
|
|
static_assert(!((ELEMENTS_PER_LATCH + 1) & ELEMENTS_PER_LATCH),
|
|
|
|
"must be one less than a power of 2");
|
|
|
|
return reinterpret_cast<page_hash_latch*>
|
|
|
|
(&array[calc_hash(fold, n) & ~ELEMENTS_PER_LATCH]);
|
|
|
|
}
|
|
|
|
public:
|
|
|
|
/** Get a page_hash latch. */
|
|
|
|
page_hash_latch *lock_get(ulint fold) const
|
|
|
|
{ return lock_get(fold, n_cells); }
|
|
|
|
|
2021-07-03 13:58:38 +03:00
|
|
|
/** Acquire an array latch.
|
2020-06-18 13:38:30 +03:00
|
|
|
@tparam exclusive whether the latch is to be acquired exclusively
|
|
|
|
@param fold hash bucket key */
|
|
|
|
template<bool exclusive> page_hash_latch *lock(ulint fold)
|
|
|
|
{
|
2021-07-03 13:58:38 +03:00
|
|
|
page_hash_latch *latch= lock_get(fold, n_cells);
|
|
|
|
latch->acquire<exclusive>();
|
|
|
|
return latch;
|
2020-06-18 13:38:30 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
/** Exclusively aqcuire all latches */
|
|
|
|
inline void write_lock_all();
|
|
|
|
|
|
|
|
/** Release all latches */
|
|
|
|
inline void write_unlock_all();
|
|
|
|
};
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Hash table of file pages (buf_page_t::in_file() holds),
|
2020-06-18 13:38:30 +03:00
|
|
|
indexed by page_id_t. Protected by both mutex and page_hash.lock_get(). */
|
|
|
|
page_hash_table page_hash;
|
MDEV-22871: Clean up hash_table_t
HASH_TABLE_SYNC_MUTEX was kind-of used for the adaptive hash index,
even though that hash table is already protected by btr_search_latches[].
HASH_TABLE_SYNC_RWLOCK was only being used for buf_pool.page_hash.
It is cleaner to decouple that synchronization from hash_table_t,
and move it to the actual user.
buf_pool_t::page_hash_latches[]: Synchronization for buf_pool.page_hash.
LATCH_ID_HASH_TABLE_MUTEX: Remove.
hash_table_t::sync_obj, hash_table_t::n_sync_obj: Remove.
hash_table_t::type, hash_table_sync_t: Remove.
HASH_ASSERT_OWN(), hash_get_mutex(), hash_get_nth_mutex(): Remove.
ib_recreate(): Merge to the only caller, buf_pool_resize_hash().
ib_create(): Merge to the callers.
ha_clear(): Merge to the only caller buf_pool_t::close().
buf_pool_t::create(): Merge the ib_create() and
hash_create_sync_obj() invocations.
ha_insert_for_fold_func(): Clarify an assertion.
buf_pool_t::page_hash_lock(): Simplify the logic.
hash_assert_can_search(), hash_assert_can_modify(): Remove.
These predicates were only being invoked for the adaptive hash index,
while they only are effective for buf_pool.page_hash.
HASH_DELETE_AND_COMPACT(): Merge to ha_delete_hash_node().
hash_get_sync_obj_index(): Remove.
hash_table_t::heaps[], hash_get_nth_heap(): Remove. It was actually unused!
hash_get_heap(): Remove. It was only used in ha_delete_hash_node(),
where we always use hash_table_t::heap.
hash_table_t::calc_hash(): Replaces hash_calc_hash().
2020-06-18 11:37:24 +03:00
|
|
|
|
2020-06-18 12:26:28 +03:00
|
|
|
/** map of block->frame to buf_block_t blocks that belong
|
|
|
|
to buf_buddy_alloc(); protected by buf_pool.mutex */
|
|
|
|
hash_table_t zip_hash;
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** number of pending read operations */
|
|
|
|
Atomic_counter<ulint> n_pend_reads;
|
2018-12-28 23:51:15 +04:00
|
|
|
Atomic_counter<ulint>
|
|
|
|
n_pend_unzip; /*!< number of pending decompressions */
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
time_t last_printout_time;
|
|
|
|
/*!< when buf_print_io was last time
|
|
|
|
called */
|
|
|
|
buf_buddy_stat_t buddy_stat[BUF_BUDDY_SIZES_MAX + 1];
|
|
|
|
/*!< Statistics of buddy system,
|
|
|
|
indexed by block size */
|
|
|
|
buf_pool_stat_t stat; /*!< current statistics */
|
|
|
|
buf_pool_stat_t old_stat; /*!< old statistics */
|
|
|
|
|
|
|
|
/* @} */
|
|
|
|
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
/** @name Page flushing algorithm fields */
|
|
|
|
/* @{ */
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
/** mutex protecting flush_list, buf_page_t::set_oldest_modification()
|
|
|
|
and buf_page_t::list pointers when !oldest_modification() */
|
2021-01-07 11:18:13 +02:00
|
|
|
MY_ALIGNED(CPU_LEVEL1_DCACHE_LINESIZE) mysql_mutex_t flush_list_mutex;
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
/** "hazard pointer" for flush_list scans; protected by flush_list_mutex */
|
|
|
|
FlushHp flush_hp;
|
|
|
|
/** modified blocks (a subset of LRU) */
|
|
|
|
UT_LIST_BASE_NODE_T(buf_page_t) flush_list;
|
2020-11-25 16:09:47 +02:00
|
|
|
private:
|
|
|
|
/** whether the page cleaner needs wakeup from indefinite sleep */
|
|
|
|
bool page_cleaner_is_idle;
|
2021-02-23 16:21:30 +08:00
|
|
|
/** track server activity count for signaling idle flushing */
|
|
|
|
ulint last_activity_count;
|
2020-11-25 16:09:47 +02:00
|
|
|
public:
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
/** signalled to wake up the page_cleaner; protected by flush_list_mutex */
|
2021-02-07 12:19:24 +02:00
|
|
|
pthread_cond_t do_flush_list;
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
|
2020-11-25 16:09:47 +02:00
|
|
|
/** @return whether the page cleaner must sleep due to being idle */
|
|
|
|
bool page_cleaner_idle() const
|
|
|
|
{
|
|
|
|
mysql_mutex_assert_owner(&flush_list_mutex);
|
|
|
|
return page_cleaner_is_idle;
|
|
|
|
}
|
|
|
|
/** Wake up the page cleaner if needed */
|
|
|
|
inline void page_cleaner_wakeup();
|
|
|
|
|
|
|
|
/** Register whether an explicit wakeup of the page cleaner is needed */
|
|
|
|
void page_cleaner_set_idle(bool deep_sleep)
|
|
|
|
{
|
|
|
|
mysql_mutex_assert_owner(&flush_list_mutex);
|
|
|
|
page_cleaner_is_idle= deep_sleep;
|
|
|
|
}
|
|
|
|
|
2021-02-23 16:21:30 +08:00
|
|
|
/** Update server last activity count */
|
|
|
|
void update_last_activity_count(ulint activity_count)
|
|
|
|
{
|
|
|
|
mysql_mutex_assert_owner(&flush_list_mutex);
|
|
|
|
last_activity_count= activity_count;
|
|
|
|
}
|
|
|
|
|
2021-06-23 13:13:11 +03:00
|
|
|
// n_flush_LRU() + n_flush_list()
|
|
|
|
// is approximately COUNT(io_fix()==BUF_IO_WRITE) in flush_list
|
2014-02-26 19:11:54 +01:00
|
|
|
|
2017-03-01 08:27:39 +02:00
|
|
|
unsigned freed_page_clock;/*!< a sequence number used
|
2014-02-26 19:11:54 +01:00
|
|
|
to count the number of buffer
|
|
|
|
blocks removed from the end of
|
|
|
|
the LRU list; NOTE that this
|
|
|
|
counter may wrap around at 4
|
|
|
|
billion! A thread is allowed
|
|
|
|
to read this for heuristic
|
|
|
|
purposes without holding any
|
|
|
|
mutex or latch */
|
2020-06-08 13:14:27 +03:00
|
|
|
bool try_LRU_scan; /*!< Cleared when an LRU
|
2014-02-26 19:11:54 +01:00
|
|
|
scan for free block fails. This
|
|
|
|
flag is used to avoid repeated
|
|
|
|
scans of LRU list when we know
|
|
|
|
that there is no free block
|
|
|
|
available in the scan depth for
|
2020-06-08 13:14:27 +03:00
|
|
|
eviction. Set whenever
|
2014-02-26 19:11:54 +01:00
|
|
|
we flush a batch from the
|
|
|
|
buffer pool. Protected by the
|
2020-03-18 21:48:00 +02:00
|
|
|
buf_pool.mutex */
|
2014-02-26 19:11:54 +01:00
|
|
|
/* @} */
|
|
|
|
|
|
|
|
/** @name LRU replacement algorithm fields */
|
|
|
|
/* @{ */
|
|
|
|
|
|
|
|
UT_LIST_BASE_NODE_T(buf_page_t) free;
|
|
|
|
/*!< base node of the free
|
|
|
|
block list */
|
2021-06-24 11:01:18 +03:00
|
|
|
/** signaled each time when the free list grows; protected by mutex */
|
|
|
|
pthread_cond_t done_free;
|
2014-11-06 13:17:11 +02:00
|
|
|
|
2016-08-12 11:17:45 +03:00
|
|
|
UT_LIST_BASE_NODE_T(buf_page_t) withdraw;
|
|
|
|
/*!< base node of the withdraw
|
|
|
|
block list. It is only used during
|
|
|
|
shrinking buffer pool size, not to
|
|
|
|
reuse the blocks will be removed */
|
|
|
|
|
|
|
|
ulint withdraw_target;/*!< target length of withdraw
|
|
|
|
block list, when withdrawing */
|
|
|
|
|
2014-11-06 13:17:11 +02:00
|
|
|
/** "hazard pointer" used during scan of LRU while doing
|
2020-03-23 09:50:35 +02:00
|
|
|
LRU list batch. Protected by buf_pool_t::mutex. */
|
2014-11-06 13:17:11 +02:00
|
|
|
LRUHp lru_hp;
|
|
|
|
|
|
|
|
/** Iterator used to scan the LRU list when searching for
|
2020-03-23 09:50:35 +02:00
|
|
|
replacable victim. Protected by buf_pool_t::mutex. */
|
2014-11-06 13:17:11 +02:00
|
|
|
LRUItr lru_scan_itr;
|
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
UT_LIST_BASE_NODE_T(buf_page_t) LRU;
|
|
|
|
/*!< base node of the LRU list */
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2014-02-26 19:11:54 +01:00
|
|
|
buf_page_t* LRU_old; /*!< pointer to the about
|
|
|
|
LRU_old_ratio/BUF_LRU_OLD_RATIO_DIV
|
|
|
|
oldest blocks in the LRU list;
|
|
|
|
NULL if LRU length less than
|
|
|
|
BUF_LRU_OLD_MIN_LEN;
|
|
|
|
NOTE: when LRU_old != NULL, its length
|
|
|
|
should always equal LRU_old_len */
|
|
|
|
ulint LRU_old_len; /*!< length of the LRU list from
|
|
|
|
the block to which LRU_old points
|
|
|
|
onward, including that block;
|
|
|
|
see buf0lru.cc for the restrictions
|
|
|
|
on this value; 0 if LRU_old == NULL;
|
|
|
|
NOTE: LRU_old_len must be adjusted
|
|
|
|
whenever LRU_old shrinks or grows! */
|
|
|
|
|
|
|
|
UT_LIST_BASE_NODE_T(buf_block_t) unzip_LRU;
|
|
|
|
/*!< base node of the
|
|
|
|
unzip_LRU list */
|
|
|
|
|
|
|
|
/* @} */
|
2020-10-15 09:49:20 +03:00
|
|
|
/** free ROW_FORMAT=COMPRESSED page frames */
|
|
|
|
UT_LIST_BASE_NODE_T(buf_buddy_free_t) zip_free[BUF_BUDDY_SIZES_MAX];
|
2019-03-21 12:05:04 +02:00
|
|
|
#if BUF_BUDDY_LOW > UNIV_ZIP_SIZE_MIN
|
|
|
|
# error "BUF_BUDDY_LOW > UNIV_ZIP_SIZE_MIN"
|
|
|
|
#endif
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Sentinels to detect if pages are read into the buffer pool while
|
|
|
|
a delete-buffering operation is pending. Protected by mutex. */
|
|
|
|
buf_page_t watch[innodb_purge_threads_MAX + 1];
|
2020-03-18 21:48:00 +02:00
|
|
|
/** Reserve a buffer. */
|
|
|
|
buf_tmp_buffer_t *io_buf_reserve() { return io_buf.reserve(); }
|
2020-06-08 13:14:27 +03:00
|
|
|
|
|
|
|
/** @return whether any I/O is pending */
|
|
|
|
bool any_io_pending() const
|
|
|
|
{
|
2021-06-23 13:13:11 +03:00
|
|
|
return n_pend_reads || n_flush_LRU() || n_flush_list();
|
2020-06-08 13:14:27 +03:00
|
|
|
}
|
|
|
|
/** @return total amount of pending I/O */
|
|
|
|
ulint io_pending() const
|
|
|
|
{
|
2021-06-23 13:13:11 +03:00
|
|
|
return n_pend_reads + n_flush_LRU() + n_flush_list();
|
2020-06-08 13:14:27 +03:00
|
|
|
}
|
2021-06-23 13:13:11 +03:00
|
|
|
|
|
|
|
private:
|
|
|
|
/** Remove a block from the flush list. */
|
|
|
|
inline void delete_from_flush_list_low(buf_page_t *bpage);
|
|
|
|
/** Remove a block from flush_list.
|
|
|
|
@param bpage buffer pool page
|
|
|
|
@param clear whether to invoke buf_page_t::clear_oldest_modification() */
|
|
|
|
void delete_from_flush_list(buf_page_t *bpage, bool clear);
|
|
|
|
public:
|
|
|
|
/** Remove a block from flush_list.
|
|
|
|
@param bpage buffer pool page */
|
|
|
|
void delete_from_flush_list(buf_page_t *bpage)
|
|
|
|
{ delete_from_flush_list(bpage, true); }
|
|
|
|
|
|
|
|
/** Insert a modified block into the flush list.
|
|
|
|
@param block modified block
|
|
|
|
@param lsn start LSN of the mini-transaction that modified the block */
|
|
|
|
void insert_into_flush_list(buf_block_t *block, lsn_t lsn);
|
|
|
|
|
|
|
|
/** Free a page whose underlying file page has been freed. */
|
|
|
|
inline void release_freed_page(buf_page_t *bpage);
|
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
private:
|
|
|
|
/** Temporary memory for page_compressed and encrypted I/O */
|
|
|
|
struct io_buf_t
|
|
|
|
{
|
|
|
|
/** number of elements in slots[] */
|
|
|
|
ulint n_slots;
|
|
|
|
/** array of slots */
|
|
|
|
buf_tmp_buffer_t *slots;
|
|
|
|
|
|
|
|
void create(ulint n_slots)
|
|
|
|
{
|
|
|
|
this->n_slots= n_slots;
|
|
|
|
slots= static_cast<buf_tmp_buffer_t*>
|
|
|
|
(ut_malloc_nokey(n_slots * sizeof *slots));
|
|
|
|
memset((void*) slots, 0, n_slots * sizeof *slots);
|
|
|
|
}
|
|
|
|
|
|
|
|
void close()
|
|
|
|
{
|
|
|
|
for (buf_tmp_buffer_t *s= slots, *e= slots + n_slots; s != e; s++)
|
|
|
|
{
|
|
|
|
aligned_free(s->crypt_buf);
|
|
|
|
aligned_free(s->comp_buf);
|
|
|
|
}
|
|
|
|
ut_free(slots);
|
|
|
|
slots= nullptr;
|
|
|
|
n_slots= 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/** Reserve a buffer */
|
|
|
|
buf_tmp_buffer_t *reserve()
|
|
|
|
{
|
|
|
|
for (buf_tmp_buffer_t *s= slots, *e= slots + n_slots; s != e; s++)
|
|
|
|
if (s->acquire())
|
|
|
|
return s;
|
|
|
|
return nullptr;
|
|
|
|
}
|
|
|
|
} io_buf;
|
|
|
|
|
|
|
|
/** whether resize() is in the critical path */
|
|
|
|
std::atomic<bool> resizing;
|
2014-02-26 19:11:54 +01:00
|
|
|
};
|
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
/** The InnoDB buffer pool */
|
|
|
|
extern buf_pool_t buf_pool;
|
|
|
|
|
2020-06-18 13:38:30 +03:00
|
|
|
inline void page_hash_latch::read_lock()
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_not_owner(&buf_pool.mutex);
|
2020-06-18 13:38:30 +03:00
|
|
|
if (!read_trylock())
|
|
|
|
read_lock_wait();
|
|
|
|
}
|
|
|
|
|
|
|
|
inline void page_hash_latch::write_lock()
|
|
|
|
{
|
|
|
|
if (!write_trylock())
|
|
|
|
write_lock_wait();
|
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline void buf_page_t::add_buf_fix_count(uint32_t count)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
buf_fix_count_+= count;
|
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline void buf_page_t::set_buf_fix_count(uint32_t count)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
buf_fix_count_= count;
|
|
|
|
}
|
2014-05-05 18:20:28 +02:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline void buf_page_t::set_state(buf_page_state state)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
2020-06-11 17:30:33 +03:00
|
|
|
#ifdef UNIV_DEBUG
|
|
|
|
switch (state) {
|
|
|
|
case BUF_BLOCK_REMOVE_HASH:
|
|
|
|
/* buf_pool_t::corrupted_evict() invokes set_corrupt_id()
|
|
|
|
before buf_LRU_free_one_page(), so we cannot assert that
|
|
|
|
we are holding the hash_lock. */
|
|
|
|
break;
|
|
|
|
case BUF_BLOCK_MEMORY:
|
|
|
|
if (!in_file()) break;
|
|
|
|
/* fall through */
|
|
|
|
case BUF_BLOCK_FILE_PAGE:
|
2020-06-18 13:38:30 +03:00
|
|
|
ut_ad(buf_pool.hash_lock_get(id_)->is_write_locked());
|
2020-06-11 17:30:33 +03:00
|
|
|
break;
|
|
|
|
case BUF_BLOCK_NOT_USED:
|
|
|
|
if (!in_file()) break;
|
|
|
|
/* fall through */
|
|
|
|
case BUF_BLOCK_ZIP_PAGE:
|
2020-06-18 13:38:30 +03:00
|
|
|
ut_ad(buf_pool.hash_lock_get(id_)->is_write_locked() ||
|
|
|
|
(this >= &buf_pool.watch[0] &&
|
|
|
|
this <= &buf_pool.watch[UT_ARR_SIZE(buf_pool.watch)]));
|
2020-06-11 17:30:33 +03:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
#endif
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
state_= state;
|
|
|
|
}
|
2014-05-05 18:20:28 +02:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline void buf_page_t::set_io_fix(buf_io_fix io_fix)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
io_fix_= io_fix;
|
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline void buf_page_t::set_corrupt_id()
|
|
|
|
{
|
|
|
|
#ifdef UNIV_DEBUG
|
2021-01-23 17:45:03 +02:00
|
|
|
switch (oldest_modification()) {
|
|
|
|
case 0:
|
|
|
|
break;
|
2021-06-23 13:13:11 +03:00
|
|
|
case 2:
|
2021-01-23 17:45:03 +02:00
|
|
|
ut_ad(fsp_is_system_temporary(id().space()));
|
2021-06-26 11:16:40 +03:00
|
|
|
/* buf_LRU_block_free_non_file_page() asserts !oldest_modification() */
|
|
|
|
ut_d(oldest_modification_= 0;)
|
2021-01-23 17:45:03 +02:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ut_ad("block is dirty" == 0);
|
|
|
|
}
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
switch (state()) {
|
|
|
|
case BUF_BLOCK_REMOVE_HASH:
|
|
|
|
break;
|
|
|
|
case BUF_BLOCK_ZIP_PAGE:
|
|
|
|
case BUF_BLOCK_FILE_PAGE:
|
2020-06-18 13:38:30 +03:00
|
|
|
ut_ad(buf_pool.hash_lock_get(id_)->is_write_locked());
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
break;
|
|
|
|
case BUF_BLOCK_NOT_USED:
|
|
|
|
case BUF_BLOCK_MEMORY:
|
|
|
|
ut_ad("invalid state" == 0);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
id_= page_id_t(~0ULL);
|
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Set oldest_modification when adding to buf_pool.flush_list */
|
|
|
|
inline void buf_page_t::set_oldest_modification(lsn_t lsn)
|
2018-02-13 23:22:20 +03:00
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.flush_list_mutex);
|
2021-06-23 13:13:11 +03:00
|
|
|
ut_ad(oldest_modification() <= 1);
|
2021-06-24 21:55:10 +03:00
|
|
|
ut_ad(lsn > 2);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
oldest_modification_= lsn;
|
2018-02-13 23:22:20 +03:00
|
|
|
}
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2021-06-24 21:55:10 +03:00
|
|
|
/** Clear oldest_modification after removing from buf_pool.flush_list */
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
inline void buf_page_t::clear_oldest_modification()
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.flush_list_mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_d(const auto state= state_);
|
|
|
|
ut_ad(state == BUF_BLOCK_FILE_PAGE || state == BUF_BLOCK_ZIP_PAGE ||
|
|
|
|
state == BUF_BLOCK_REMOVE_HASH);
|
|
|
|
ut_ad(oldest_modification());
|
2021-06-26 11:16:40 +03:00
|
|
|
ut_ad(!list.prev);
|
|
|
|
ut_ad(!list.next);
|
|
|
|
/* We must use release memory order to guarantee that callers of
|
|
|
|
oldest_modification_acquire() will observe the block as
|
|
|
|
being detached from buf_pool.flush_list, after reading the value 0. */
|
|
|
|
oldest_modification_.store(0, std::memory_order_release);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
|
2021-06-23 13:13:11 +03:00
|
|
|
/** Note that a block is no longer dirty, while not removing
|
|
|
|
it from buf_pool.flush_list */
|
|
|
|
inline void buf_page_t::clear_oldest_modification(bool temporary)
|
|
|
|
{
|
|
|
|
ut_ad(temporary == fsp_is_system_temporary(id().space()));
|
2021-06-26 11:16:40 +03:00
|
|
|
if (temporary)
|
|
|
|
{
|
|
|
|
ut_ad(oldest_modification() == 2);
|
|
|
|
oldest_modification_= 0;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
/* We use release memory order to guarantee that callers of
|
|
|
|
oldest_modification_acquire() will observe the block as
|
|
|
|
being detached from buf_pool.flush_list, after reading the value 0. */
|
|
|
|
ut_ad(oldest_modification() > 2);
|
|
|
|
oldest_modification_.store(1, std::memory_order_release);
|
|
|
|
}
|
2021-06-23 13:13:11 +03:00
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** @return whether the block is modified and ready for flushing */
|
|
|
|
inline bool buf_page_t::ready_for_flush() const
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(in_LRU_list);
|
|
|
|
ut_a(in_file());
|
2021-06-23 13:13:11 +03:00
|
|
|
ut_ad(fsp_is_system_temporary(id().space())
|
|
|
|
? oldest_modification() == 2
|
|
|
|
: oldest_modification() > 2);
|
|
|
|
return io_fix_ == BUF_IO_NONE;
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
}
|
2016-08-12 11:17:45 +03:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** @return whether the block can be relocated in memory.
|
|
|
|
The block can be dirty, but it must not be I/O-fixed or bufferfixed. */
|
|
|
|
inline bool buf_page_t::can_relocate() const
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(in_file());
|
|
|
|
ut_ad(in_LRU_list);
|
|
|
|
return io_fix_ == BUF_IO_NONE && !buf_fix_count_;
|
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** @return whether the block has been flagged old in buf_pool.LRU */
|
|
|
|
inline bool buf_page_t::is_old() const
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(in_file());
|
|
|
|
ut_ad(in_LRU_list);
|
|
|
|
return old;
|
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
/** Set whether a block is old in buf_pool.LRU */
|
|
|
|
inline void buf_page_t::set_old(bool old)
|
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex);
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
ut_ad(in_LRU_list);
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
#ifdef UNIV_LRU_DEBUG
|
|
|
|
ut_a((buf_pool.LRU_old_len == 0) == (buf_pool.LRU_old == nullptr));
|
|
|
|
/* If a block is flagged "old", the LRU_old list must exist. */
|
|
|
|
ut_a(!old || buf_pool.LRU_old);
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
if (UT_LIST_GET_PREV(LRU, this) && UT_LIST_GET_NEXT(LRU, this))
|
|
|
|
{
|
|
|
|
const buf_page_t *prev= UT_LIST_GET_PREV(LRU, this);
|
|
|
|
const buf_page_t *next = UT_LIST_GET_NEXT(LRU, this);
|
|
|
|
if (prev->old == next->old)
|
|
|
|
ut_a(prev->old == old);
|
|
|
|
else
|
|
|
|
{
|
|
|
|
ut_a(!prev->old);
|
|
|
|
ut_a(buf_pool.LRU_old == (old ? this : next));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif /* UNIV_LRU_DEBUG */
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
this->old= old;
|
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
#ifdef UNIV_DEBUG
|
2014-02-26 19:11:54 +01:00
|
|
|
/** Forbid the release of the buffer pool mutex. */
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
# define buf_pool_mutex_exit_forbid() do { \
|
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex); \
|
|
|
|
buf_pool.mutex_exit_forbidden++; \
|
2014-02-26 19:11:54 +01:00
|
|
|
} while (0)
|
|
|
|
/** Allow the release of the buffer pool mutex. */
|
2020-02-12 14:45:21 +02:00
|
|
|
# define buf_pool_mutex_exit_allow() do { \
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(&buf_pool.mutex); \
|
|
|
|
ut_ad(buf_pool.mutex_exit_forbidden--); \
|
2014-02-26 19:11:54 +01:00
|
|
|
} while (0)
|
|
|
|
#else
|
|
|
|
/** Forbid the release of the buffer pool mutex. */
|
2020-02-12 14:45:21 +02:00
|
|
|
# define buf_pool_mutex_exit_forbid() ((void) 0)
|
2014-02-26 19:11:54 +01:00
|
|
|
/** Allow the release of the buffer pool mutex. */
|
2020-02-12 14:45:21 +02:00
|
|
|
# define buf_pool_mutex_exit_allow() ((void) 0)
|
2014-02-26 19:11:54 +01:00
|
|
|
#endif
|
|
|
|
|
|
|
|
/**********************************************************************
|
|
|
|
Let us list the consistency conditions for different control block states.
|
|
|
|
|
|
|
|
NOT_USED: is in free list, not in LRU list, not in flush list, nor
|
|
|
|
page hash table
|
|
|
|
MEMORY: is not in free list, LRU list, or flush list, nor page
|
|
|
|
hash table
|
|
|
|
FILE_PAGE: space and offset are defined, is in page hash table
|
|
|
|
if io_fix == BUF_IO_WRITE,
|
2021-06-23 13:13:11 +03:00
|
|
|
buf_pool.n_flush_LRU() || buf_pool.n_flush_list()
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
(1) if buf_fix_count == 0, then
|
|
|
|
is in LRU list, not in free list
|
|
|
|
is in flush list,
|
|
|
|
if and only if oldest_modification > 0
|
|
|
|
is x-locked,
|
|
|
|
if and only if io_fix == BUF_IO_READ
|
|
|
|
is s-locked,
|
|
|
|
if and only if io_fix == BUF_IO_WRITE
|
|
|
|
|
|
|
|
(2) if buf_fix_count > 0, then
|
|
|
|
is not in LRU list, not in free list
|
|
|
|
is in flush list,
|
|
|
|
if and only if oldest_modification > 0
|
|
|
|
if io_fix == BUF_IO_READ,
|
|
|
|
is x-locked
|
|
|
|
if io_fix == BUF_IO_WRITE,
|
|
|
|
is s-locked
|
|
|
|
|
|
|
|
State transitions:
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
NOT_USED => MEMORY
|
|
|
|
MEMORY => FILE_PAGE
|
2014-02-26 19:11:54 +01:00
|
|
|
MEMORY => NOT_USED
|
|
|
|
FILE_PAGE => NOT_USED NOTE: This transition is allowed if and only if
|
|
|
|
(1) buf_fix_count == 0,
|
|
|
|
(2) oldest_modification == 0, and
|
|
|
|
(3) io_fix == 0.
|
|
|
|
*/
|
|
|
|
|
2020-02-12 14:45:21 +02:00
|
|
|
/** Select from where to start a scan. If we have scanned
|
|
|
|
too deep into the LRU list it resets the value to the tail
|
|
|
|
of the LRU list.
|
|
|
|
@return buf_page_t from where to start scan. */
|
2020-03-18 21:48:00 +02:00
|
|
|
inline buf_page_t *LRUItr::start()
|
2020-02-12 14:45:21 +02:00
|
|
|
{
|
MDEV-23399: Performance regression with write workloads
The buffer pool refactoring in MDEV-15053 and MDEV-22871 shifted
the performance bottleneck to the page flushing.
The configuration parameters will be changed as follows:
innodb_lru_flush_size=32 (new: how many pages to flush on LRU eviction)
innodb_lru_scan_depth=1536 (old: 1024)
innodb_max_dirty_pages_pct=90 (old: 75)
innodb_max_dirty_pages_pct_lwm=75 (old: 0)
Note: The parameter innodb_lru_scan_depth will only affect LRU
eviction of buffer pool pages when a new page is being allocated. The
page cleaner thread will no longer evict any pages. It used to
guarantee that some pages will remain free in the buffer pool. Now, we
perform that eviction 'on demand' in buf_LRU_get_free_block().
The parameter innodb_lru_scan_depth(srv_LRU_scan_depth) is used as follows:
* When the buffer pool is being shrunk in buf_pool_t::withdraw_blocks()
* As a buf_pool.free limit in buf_LRU_list_batch() for terminating
the flushing that is initiated e.g., by buf_LRU_get_free_block()
The parameter also used to serve as an initial limit for unzip_LRU
eviction (evicting uncompressed page frames while retaining
ROW_FORMAT=COMPRESSED pages), but now we will use a hard-coded limit
of 100 or unlimited for invoking buf_LRU_scan_and_free_block().
The status variables will be changed as follows:
innodb_buffer_pool_pages_flushed: This includes also the count of
innodb_buffer_pool_pages_LRU_flushed and should work reliably,
updated one by one in buf_flush_page() to give more real-time
statistics. The function buf_flush_stats(), which we are removing,
was not called in every code path. For both counters, we will use
regular variables that are incremented in a critical section of
buf_pool.mutex. Note that show_innodb_vars() directly links to the
variables, and reads of the counters will *not* be protected by
buf_pool.mutex, so you cannot get a consistent snapshot of both variables.
The following INFORMATION_SCHEMA.INNODB_METRICS counters will be
removed, because the page cleaner no longer deals with writing or
evicting least recently used pages, and because the single-page writes
have been removed:
* buffer_LRU_batch_flush_avg_time_slot
* buffer_LRU_batch_flush_avg_time_thread
* buffer_LRU_batch_flush_avg_time_est
* buffer_LRU_batch_flush_avg_pass
* buffer_LRU_single_flush_scanned
* buffer_LRU_single_flush_num_scan
* buffer_LRU_single_flush_scanned_per_call
When moving to a single buffer pool instance in MDEV-15058, we missed
some opportunity to simplify the buf_flush_page_cleaner thread. It was
unnecessarily using a mutex and some complex data structures, even
though we always have a single page cleaner thread.
Furthermore, the buf_flush_page_cleaner thread had separate 'recovery'
and 'shutdown' modes where it was waiting to be triggered by some
other thread, adding unnecessary latency and potential for hangs in
relatively rarely executed startup or shutdown code.
The page cleaner was also running two kinds of batches in an
interleaved fashion: "LRU flush" (writing out some least recently used
pages and evicting them on write completion) and the normal batches
that aim to increase the MIN(oldest_modification) in the buffer pool,
to help the log checkpoint advance.
The buf_pool.flush_list flushing was being blocked by
buf_block_t::lock for no good reason. Furthermore, if the FIL_PAGE_LSN
of a page is ahead of log_sys.get_flushed_lsn(), that is, what has
been persistently written to the redo log, we would trigger a log
flush and then resume the page flushing. This would unnecessarily
limit the performance of the page cleaner thread and trigger the
infamous messages "InnoDB: page_cleaner: 1000ms intended loop took 4450ms.
The settings might not be optimal" that were suppressed in
commit d1ab89037a518fcffbc50c24e4bd94e4ec33aed0 unless log_warnings>2.
Our revised algorithm will make log_sys.get_flushed_lsn() advance at
the start of buf_flush_lists(), and then execute a 'best effort' to
write out all pages. The flush batches will skip pages that were modified
since the log was written, or are are currently exclusively locked.
The MDEV-13670 message "page_cleaner: 1000ms intended loop took" message
will be removed, because by design, the buf_flush_page_cleaner() should
not be blocked during a batch for extended periods of time.
We will remove the single-page flushing altogether. Related to this,
the debug parameter innodb_doublewrite_batch_size will be removed,
because all of the doublewrite buffer will be used for flushing
batches. If a page needs to be evicted from the buffer pool and all
100 least recently used pages in the buffer pool have unflushed
changes, buf_LRU_get_free_block() will execute buf_flush_lists() to
write out and evict innodb_lru_flush_size pages. At most one thread
will execute buf_flush_lists() in buf_LRU_get_free_block(); other
threads will wait for that LRU flushing batch to finish.
To improve concurrency, we will replace the InnoDB ib_mutex_t and
os_event_t native mutexes and condition variables in this area of code.
Most notably, this means that the buffer pool mutex (buf_pool.mutex)
is no longer instrumented via any InnoDB interfaces. It will continue
to be instrumented via PERFORMANCE_SCHEMA.
For now, both buf_pool.flush_list_mutex and buf_pool.mutex will be
declared with MY_MUTEX_INIT_FAST (PTHREAD_MUTEX_ADAPTIVE_NP). The critical
sections of buf_pool.flush_list_mutex should be shorter than those for
buf_pool.mutex, because in the worst case, they cover a linear scan of
buf_pool.flush_list, while the worst case of a critical section of
buf_pool.mutex covers a linear scan of the potentially much longer
buf_pool.LRU list.
mysql_mutex_is_owner(), safe_mutex_is_owner(): New predicate, usable
with SAFE_MUTEX. Some InnoDB debug assertions need this predicate
instead of mysql_mutex_assert_owner() or mysql_mutex_assert_not_owner().
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaces buf_pool_t::init_flush[] and buf_pool_t::n_flush[].
The number of active flush operations.
buf_pool_t::mutex, buf_pool_t::flush_list_mutex: Use mysql_mutex_t
instead of ib_mutex_t, to have native mutexes with PERFORMANCE_SCHEMA
and SAFE_MUTEX instrumentation.
buf_pool_t::done_flush_LRU: Condition variable for !n_flush_LRU.
buf_pool_t::done_flush_list: Condition variable for !n_flush_list.
buf_pool_t::do_flush_list: Condition variable to wake up the
buf_flush_page_cleaner when a log checkpoint needs to be written
or the server is being shut down. Replaces buf_flush_event.
We will keep using timed waits (the page cleaner thread will wake
_at least_ once per second), because the calculations for
innodb_adaptive_flushing depend on fixed time intervals.
buf_dblwr: Allocate statically, and move all code to member functions.
Use a native mutex and condition variable. Remove code to deal with
single-page flushing.
buf_dblwr_check_block(): Make the check debug-only. We were spending
a significant amount of execution time in page_simple_validate_new().
flush_counters_t::unzip_LRU_evicted: Remove.
IORequest: Make more members const. FIXME: m_fil_node should be removed.
buf_flush_sync_lsn: Protect by std::atomic, not page_cleaner.mutex
(which we are removing).
page_cleaner_slot_t, page_cleaner_t: Remove many redundant members.
pc_request_flush_slot(): Replaces pc_request() and pc_flush_slot().
recv_writer_thread: Remove. Recovery works just fine without it, if we
simply invoke buf_flush_sync() at the end of each batch in
recv_sys_t::apply().
recv_recovery_from_checkpoint_finish(): Remove. We can simply call
recv_sys.debug_free() directly.
srv_started_redo: Replaces srv_start_state.
SRV_SHUTDOWN_FLUSH_PHASE: Remove. logs_empty_and_mark_files_at_shutdown()
can communicate with the normal page cleaner loop via the new function
flush_buffer_pool().
buf_flush_remove(): Assert that the calling thread is holding
buf_pool.flush_list_mutex. This removes unnecessary mutex operations
from buf_flush_remove_pages() and buf_flush_dirty_pages(),
which replace buf_LRU_flush_or_remove_pages().
buf_flush_lists(): Renamed from buf_flush_batch(), with simplified
interface. Return the number of flushed pages. Clarified comments and
renamed min_n to max_n. Identify LRU batch by lsn=0. Merge all the functions
buf_flush_start(), buf_flush_batch(), buf_flush_end() directly to this
function, which was their only caller, and remove 2 unnecessary
buf_pool.mutex release/re-acquisition that we used to perform around
the buf_flush_batch() call. At the start, if not all log has been
durably written, wait for a background task to do it, or start a new
task to do it. This allows the log write to run concurrently with our
page flushing batch. Any pages that were skipped due to too recent
FIL_PAGE_LSN or due to them being latched by a writer should be flushed
during the next batch, unless there are further modifications to those
pages. It is possible that a page that we must flush due to small
oldest_modification also carries a recent FIL_PAGE_LSN or is being
constantly modified. In the worst case, all writers would then end up
waiting in log_free_check() to allow the flushing and the checkpoint
to complete.
buf_do_flush_list_batch(): Clarify comments, and rename min_n to max_n.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_flush_space(): Auxiliary function to look up a tablespace for
page flushing.
buf_flush_page(): Defer the computation of space->full_crc32(). Never
call log_write_up_to(), but instead skip persistent pages whose latest
modification (FIL_PAGE_LSN) is newer than the redo log. Also skip
pages on which we cannot acquire a shared latch without waiting.
buf_flush_try_neighbors(): Do not bother checking buf_fix_count
because buf_flush_page() will no longer wait for the page latch.
Take the tablespace as a parameter, and only execute this function
when innodb_flush_neighbors>0. Avoid repeated calls of page_id_t::fold().
buf_flush_relocate_on_flush_list(): Declare as cold, and push down
a condition from the callers.
buf_flush_check_neighbor(): Take id.fold() as a parameter.
buf_flush_sync(): Ensure that the buf_pool.flush_list is empty,
because the flushing batch will skip pages whose modifications have
not yet been written to the log or were latched for modification.
buf_free_from_unzip_LRU_list_batch(): Remove redundant local variables.
buf_flush_LRU_list_batch(): Let the caller buf_do_LRU_batch() initialize
the counters, and report n->evicted.
Cache the last looked up tablespace. If neighbor flushing is not applicable,
invoke buf_flush_page() directly, avoiding a page lookup in between.
buf_do_LRU_batch(): Return the number of pages flushed.
buf_LRU_free_page(): Only release and re-acquire buf_pool.mutex if
adaptive hash index entries are pointing to the block.
buf_LRU_get_free_block(): Do not wake up the page cleaner, because it
will no longer perform any useful work for us, and we do not want it
to compete for I/O while buf_flush_lists(innodb_lru_flush_size, 0)
writes out and evicts at most innodb_lru_flush_size pages. (The
function buf_do_LRU_batch() may complete after writing fewer pages if
more than innodb_lru_scan_depth pages end up in buf_pool.free list.)
Eliminate some mutex release-acquire cycles, and wait for the LRU
flush batch to complete before rescanning.
buf_LRU_check_size_of_non_data_objects(): Simplify the code.
buf_page_write_complete(): Remove the parameter evict, and always
evict pages that were part of an LRU flush.
buf_page_create(): Take a pre-allocated page as a parameter.
buf_pool_t::free_block(): Free a pre-allocated block.
recv_sys_t::recover_low(), recv_sys_t::apply(): Preallocate the block
while not holding recv_sys.mutex. During page allocation, we may
initiate a page flush, which in turn may initiate a log flush, which
would require acquiring log_sys.mutex, which should always be acquired
before recv_sys.mutex in order to avoid deadlocks. Therefore, we must
not be holding recv_sys.mutex while allocating a buffer pool block.
BtrBulk::logFreeCheck(): Skip a redundant condition.
row_undo_step(): Do not invoke srv_inc_activity_count() for every row
that is being rolled back. It should suffice to invoke the function in
trx_flush_log_if_needed() during trx_t::commit_in_memory() when the
rollback completes.
sync_check_enable(): Remove. We will enable innodb_sync_debug from the
very beginning.
Reviewed by: Vladislav Vaintroub
2020-10-15 12:10:42 +03:00
|
|
|
mysql_mutex_assert_owner(m_mutex);
|
2020-02-12 14:45:21 +02:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
if (!m_hp || m_hp->old)
|
|
|
|
m_hp= UT_LIST_GET_LAST(buf_pool.LRU);
|
2020-02-12 14:45:21 +02:00
|
|
|
|
2020-03-18 21:48:00 +02:00
|
|
|
return m_hp;
|
2020-02-12 14:45:21 +02:00
|
|
|
}
|
|
|
|
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
#ifdef UNIV_DEBUG
|
2014-02-26 19:11:54 +01:00
|
|
|
/** Functor to validate the LRU list. */
|
|
|
|
struct CheckInLRUList {
|
|
|
|
void operator()(const buf_page_t* elem) const
|
|
|
|
{
|
|
|
|
ut_a(elem->in_LRU_list);
|
|
|
|
}
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2020-02-12 14:45:21 +02:00
|
|
|
static void validate()
|
2016-08-12 11:17:45 +03:00
|
|
|
{
|
2020-03-18 21:48:00 +02:00
|
|
|
ut_list_validate(buf_pool.LRU, CheckInLRUList());
|
2016-08-12 11:17:45 +03:00
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
};
|
|
|
|
|
|
|
|
/** Functor to validate the LRU list. */
|
|
|
|
struct CheckInFreeList {
|
|
|
|
void operator()(const buf_page_t* elem) const
|
|
|
|
{
|
|
|
|
ut_a(elem->in_free_list);
|
|
|
|
}
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2020-02-12 14:45:21 +02:00
|
|
|
static void validate()
|
2016-08-12 11:17:45 +03:00
|
|
|
{
|
2020-03-18 21:48:00 +02:00
|
|
|
ut_list_validate(buf_pool.free, CheckInFreeList());
|
2016-08-12 11:17:45 +03:00
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
};
|
|
|
|
|
|
|
|
struct CheckUnzipLRUAndLRUList {
|
|
|
|
void operator()(const buf_block_t* elem) const
|
|
|
|
{
|
|
|
|
ut_a(elem->page.in_LRU_list);
|
|
|
|
ut_a(elem->in_unzip_LRU_list);
|
|
|
|
}
|
2016-08-12 11:17:45 +03:00
|
|
|
|
2020-02-12 14:45:21 +02:00
|
|
|
static void validate()
|
2016-08-12 11:17:45 +03:00
|
|
|
{
|
2020-03-18 21:48:00 +02:00
|
|
|
ut_list_validate(buf_pool.unzip_LRU,
|
2019-04-29 14:33:46 +03:00
|
|
|
CheckUnzipLRUAndLRUList());
|
2016-08-12 11:17:45 +03:00
|
|
|
}
|
2014-02-26 19:11:54 +01:00
|
|
|
};
|
MDEV-15053 Reduce buf_pool_t::mutex contention
User-visible changes: The INFORMATION_SCHEMA views INNODB_BUFFER_PAGE
and INNODB_BUFFER_PAGE_LRU will report a dummy value FLUSH_TYPE=0
and will no longer report the PAGE_STATE value READY_FOR_USE.
We will remove some fields from buf_page_t and move much code to
member functions of buf_pool_t and buf_page_t, so that the access
rules of data members can be enforced consistently.
Evicting or adding pages in buf_pool.LRU will remain covered by
buf_pool.mutex.
Evicting or adding pages in buf_pool.page_hash will remain
covered by both buf_pool.mutex and the buf_pool.page_hash X-latch.
After this fix, buf_pool.page_hash lookups can entirely
avoid acquiring buf_pool.mutex, only relying on
buf_pool.hash_lock_get() S-latch.
Similarly, buf_flush_check_neighbors() can will rely solely on
buf_pool.mutex, no buf_pool.page_hash latch at all.
The buf_pool.mutex is rather contended in I/O heavy benchmarks,
especially when the workload does not fit in the buffer pool.
The first attempt to alleviate the contention was the
buf_pool_t::mutex split in
commit 4ed7082eefe56b3e97e0edefb3df76dd7ef5e858
which introduced buf_block_t::mutex, which we are now removing.
Later, multiple instances of buf_pool_t were introduced
in commit c18084f71b02ea707c6461353e6cfc15d7553bc6
and recently removed by us in
commit 1a6f708ec594ac0ae2dd30db926ab07b100fa24b (MDEV-15058).
UNIV_BUF_DEBUG: Remove. This option to enable some buffer pool
related debugging in otherwise non-debug builds has not been used
for years. Instead, we have been using UNIV_DEBUG, which is enabled
in CMAKE_BUILD_TYPE=Debug.
buf_block_t::mutex, buf_pool_t::zip_mutex: Remove. We can mainly rely on
std::atomic and the buf_pool.page_hash latches, and in some cases
depend on buf_pool.mutex or buf_pool.flush_list_mutex just like before.
We must always release buf_block_t::lock before invoking
unfix() or io_unfix(), to prevent a glitch where a block that was
added to the buf_pool.free list would apper X-latched. See
commit c5883debd6ef440a037011c11873b396923e93c5 how this glitch
was finally caught in a debug environment.
We move some buf_pool_t::page_hash specific code from the
ha and hash modules to buf_pool, for improved readability.
buf_pool_t::close(): Assert that all blocks are clean, except
on aborted startup or crash-like shutdown.
buf_pool_t::validate(): No longer attempt to validate
n_flush[] against the number of BUF_IO_WRITE fixed blocks,
because buf_page_t::flush_type no longer exists.
buf_pool_t::watch_set(): Replaces buf_pool_watch_set().
Reduce mutex contention by separating the buf_pool.watch[]
allocation and the insert into buf_pool.page_hash.
buf_pool_t::page_hash_lock<bool exclusive>(): Acquire a
buf_pool.page_hash latch.
Replaces and extends buf_page_hash_lock_s_confirm()
and buf_page_hash_lock_x_confirm().
buf_pool_t::READ_AHEAD_PAGES: Renamed from BUF_READ_AHEAD_PAGES.
buf_pool_t::curr_size, old_size, read_ahead_area, n_pend_reads:
Use Atomic_counter.
buf_pool_t::running_out(): Replaces buf_LRU_buf_pool_running_out().
buf_pool_t::LRU_remove(): Remove a block from the LRU list
and return its predecessor. Incorporates buf_LRU_adjust_hp(),
which was removed.
buf_page_get_gen(): Remove a redundant call of fsp_is_system_temporary(),
for mode == BUF_GET_IF_IN_POOL_OR_WATCH, which is only used by
BTR_DELETE_OP (purge), which is never invoked on temporary tables.
buf_free_from_unzip_LRU_list_batch(): Avoid redundant assignments.
buf_LRU_free_from_unzip_LRU_list(): Simplify the loop condition.
buf_LRU_free_page(): Clarify the function comment.
buf_flush_check_neighbor(), buf_flush_check_neighbors():
Rewrite the construction of the page hash range. We will hold
the buf_pool.mutex for up to buf_pool.read_ahead_area (at most 64)
consecutive lookups of buf_pool.page_hash.
buf_flush_page_and_try_neighbors(): Remove.
Merge to its only callers, and remove redundant operations in
buf_flush_LRU_list_batch().
buf_read_ahead_random(), buf_read_ahead_linear(): Rewrite.
Do not acquire buf_pool.mutex, and iterate directly with page_id_t.
ut_2_power_up(): Remove. my_round_up_to_next_power() is inlined
and avoids any loops.
fil_page_get_prev(), fil_page_get_next(), fil_addr_is_null(): Remove.
buf_flush_page(): Add a fil_space_t* parameter. Minimize the
buf_pool.mutex hold time. buf_pool.n_flush[] is no longer updated
atomically with the io_fix, and we will protect most buf_block_t
fields with buf_block_t::lock. The function
buf_flush_write_block_low() is removed and merged here.
buf_page_init_for_read(): Use static linkage. Initialize the newly
allocated block and acquire the exclusive buf_block_t::lock while not
holding any mutex.
IORequest::IORequest(): Remove the body. We only need to invoke
set_punch_hole() in buf_flush_page() and nowhere else.
buf_page_t::flush_type: Remove. Replaced by IORequest::flush_type.
This field is only used during a fil_io() call.
That function already takes IORequest as a parameter, so we had
better introduce for the rarely changing field.
buf_block_t::init(): Replaces buf_page_init().
buf_page_t::init(): Replaces buf_page_init_low().
buf_block_t::initialise(): Initialise many fields, but
keep the buf_page_t::state(). Both buf_pool_t::validate() and
buf_page_optimistic_get() requires that buf_page_t::in_file()
be protected atomically with buf_page_t::in_page_hash
and buf_page_t::in_LRU_list.
buf_page_optimistic_get(): Now that buf_block_t::mutex
no longer exists, we must check buf_page_t::io_fix()
after acquiring the buf_pool.page_hash lock, to detect
whether buf_page_init_for_read() has been initiated.
We will also check the io_fix() before acquiring hash_lock
in order to avoid unnecessary computation.
The field buf_block_t::modify_clock (protected by buf_block_t::lock)
allows buf_page_optimistic_get() to validate the block.
buf_page_t::real_size: Remove. It was only used while flushing
pages of page_compressed tables.
buf_page_encrypt(): Add an output parameter that allows us ot eliminate
buf_page_t::real_size. Replace a condition with debug assertion.
buf_page_should_punch_hole(): Remove.
buf_dblwr_t::add_to_batch(): Replaces buf_dblwr_add_to_batch().
Add the parameter size (to replace buf_page_t::real_size).
buf_dblwr_t::write_single_page(): Replaces buf_dblwr_write_single_page().
Add the parameter size (to replace buf_page_t::real_size).
fil_system_t::detach(): Replaces fil_space_detach().
Ensure that fil_validate() will not be violated even if
fil_system.mutex is released and reacquired.
fil_node_t::complete_io(): Renamed from fil_node_complete_io().
fil_node_t::close_to_free(): Replaces fil_node_close_to_free().
Avoid invoking fil_node_t::close() because fil_system.n_open
has already been decremented in fil_space_t::detach().
BUF_BLOCK_READY_FOR_USE: Remove. Directly use BUF_BLOCK_MEMORY.
BUF_BLOCK_ZIP_DIRTY: Remove. Directly use BUF_BLOCK_ZIP_PAGE,
and distinguish dirty pages by buf_page_t::oldest_modification().
BUF_BLOCK_POOL_WATCH: Remove. Use BUF_BLOCK_NOT_USED instead.
This state was only being used for buf_page_t that are in
buf_pool.watch.
buf_pool_t::watch[]: Remove pointer indirection.
buf_page_t::in_flush_list: Remove. It was set if and only if
buf_page_t::oldest_modification() is nonzero.
buf_page_decrypt_after_read(), buf_corrupt_page_release(),
buf_page_check_corrupt(): Change the const fil_space_t* parameter
to const fil_node_t& so that we can report the correct file name.
buf_page_monitor(): Declare as an ATTRIBUTE_COLD global function.
buf_page_io_complete(): Split to buf_page_read_complete() and
buf_page_write_complete().
buf_dblwr_t::in_use: Remove.
buf_dblwr_t::buf_block_array: Add IORequest::flush_t.
buf_dblwr_sync_datafiles(): Remove. It was a useless wrapper of
os_aio_wait_until_no_pending_writes().
buf_flush_write_complete(): Declare static, not global.
Add the parameter IORequest::flush_t.
buf_flush_freed_page(): Simplify the code.
recv_sys_t::flush_lru: Renamed from flush_type and changed to bool.
fil_read(), fil_write(): Replaced with direct use of fil_io().
fil_buffering_disabled(): Remove. Check srv_file_flush_method directly.
fil_mutex_enter_and_prepare_for_io(): Return the resolved
fil_space_t* to avoid a duplicated lookup in the caller.
fil_report_invalid_page_access(): Clean up the parameters.
fil_io(): Return fil_io_t, which comprises fil_node_t and error code.
Always invoke fil_space_t::acquire_for_io() and let either the
sync=true caller or fil_aio_callback() invoke
fil_space_t::release_for_io().
fil_aio_callback(): Rewrite to replace buf_page_io_complete().
fil_check_pending_operations(): Remove a parameter, and remove some
redundant lookups.
fil_node_close_to_free(): Wait for n_pending==0. Because we no longer
do an extra lookup of the tablespace between fil_io() and the
completion of the operation, we must give fil_node_t::complete_io() a
chance to decrement the counter.
fil_close_tablespace(): Remove unused parameter trx, and document
that this is only invoked during the error handling of IMPORT TABLESPACE.
row_import_discard_changes(): Merged with the only caller,
row_import_cleanup(). Do not lock up the data dictionary while
invoking fil_close_tablespace().
logs_empty_and_mark_files_at_shutdown(): Do not invoke
fil_close_all_files(), to avoid a !needs_flush assertion failure
on fil_node_t::close().
innodb_shutdown(): Invoke os_aio_free() before fil_close_all_files().
fil_close_all_files(): Invoke fil_flush_file_spaces()
to ensure proper durability.
thread_pool::unbind(): Fix a crash that would occur on Windows
after srv_thread_pool->disable_aio() and os_file_close().
This fix was submitted by Vladislav Vaintroub.
Thanks to Matthias Leich and Axel Schwenke for extensive testing,
Vladislav Vaintroub for helpful comments, and Eugene Kosov for a review.
2020-06-05 12:35:46 +03:00
|
|
|
#endif /* UNIV_DEBUG */
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
#include "buf0buf.ic"
|
2016-04-22 10:50:45 +02:00
|
|
|
|
2016-08-12 11:17:45 +03:00
|
|
|
#endif /* !UNIV_INNOCHECKSUM */
|
2014-02-26 19:11:54 +01:00
|
|
|
|
|
|
|
#endif
|