buf_LRU_block_remove_hashed(): Ever since
commit 2e814d4702
we could get page_zip_validate() failures after an ALTER TABLE
operation was aborted and BtrBulk::pageCommit() had never been
executed on some blocks.
For more convenient monitoring of something that could greatly affect
the volume of page writes, we add the status variable
Innodb_buffer_pool_pages_split that was previously only available
via information_schema.innodb_metrics as "innodb_page_splits".
This was suggested by Axel Schwenke.
buf_flush_page_count: Replaced with buf_pool.stat.n_pages_written.
We protect buf_pool.stat (except n_page_gets) with buf_pool.mutex
and remove unnecessary export_vars indirection.
buf_pool.flush_list_bytes: Moved from buf_pool.stat.flush_list_bytes.
Protected by buf_pool.flush_list_mutex.
buf_pool_t::page_cleaner_status: Replaces buf_pool_t::n_flush_LRU_,
buf_pool_t::n_flush_list_, and buf_pool_t::page_cleaner_is_idle.
Protected by buf_pool.flush_list_mutex. We will exclusively broadcast
buf_pool.done_flush_list by the buf_flush_page_cleaner thread,
and only wait for it when communicating with buf_flush_page_cleaner.
There is no need to keep a count of pending writes by the
buf_pool.flush_list processing. A single flag suffices for that.
Waits for page write completion can be performed by
simply waiting on block->page.lock, or by invoking
buf_dblwr.wait_for_page_writes().
buf_LRU_block_free_non_file_page(): Broadcast buf_pool.done_free and
set buf_pool.try_LRU_scan when freeing a page. This would be
executed also as part of buf_page_write_complete().
buf_page_write_complete(): Do not broadcast buf_pool.done_flush_list,
and do not acquire buf_pool.mutex unless buf_pool.LRU eviction is needed.
Let buf_dblwr count all writes to persistent pages and broadcast a
condition variable when no outstanding writes remain.
buf_flush_page_cleaner(): Prioritize LRU flushing and eviction right after
"furious flushing" (lsn_limit). Simplify the conditions and reduce the
hold time of buf_pool.flush_list_mutex. Refuse to shut down
or sleep if buf_pool.ran_out(), that is, LRU eviction is needed.
buf_pool_t::page_cleaner_wakeup(): Add the optional parameter for_LRU.
buf_LRU_get_free_block(): Protect buf_lru_free_blocks_error_printed
with buf_pool.mutex. Invoke buf_pool.page_cleaner_wakeup(true) to
to ensure that buf_flush_page_cleaner() will process the LRU flush
request.
buf_do_LRU_batch(), buf_flush_list(), buf_flush_list_space():
Update buf_pool.stat.n_pages_written when submitting writes
(while holding buf_pool.mutex), not when completing them.
buf_page_t::flush(), buf_flush_discard_page(): Require that
the page U-latch be acquired upfront, and remove
buf_page_t::ready_for_flush().
buf_pool_t::delete_from_flush_list(): Remove the parameter "bool clear".
buf_flush_page(): Count pending page writes via buf_dblwr.
buf_flush_try_neighbors(): Take the block of page_id as a parameter.
If the tablespace is dropped before our page has been written out,
release the page U-latch.
buf_pool_invalidate(): Let the caller ensure that there are no
outstanding writes.
buf_flush_wait_batch_end(false),
buf_flush_wait_batch_end_acquiring_mutex(false):
Replaced with buf_dblwr.wait_for_page_writes().
buf_flush_wait_LRU_batch_end(): Replaces buf_flush_wait_batch_end(true).
buf_flush_list(): Remove some broadcast of buf_pool.done_flush_list.
buf_flush_buffer_pool(): Invoke also buf_dblwr.wait_for_page_writes().
buf_pool_t::io_pending(), buf_pool_t::n_flush_list(): Remove.
Outstanding writes are reflected by buf_dblwr.pending_writes().
buf_dblwr_t::init(): New function, to initialize the mutex and
the condition variables, but not the backing store.
buf_dblwr_t::is_created(): Replaces buf_dblwr_t::is_initialised().
buf_dblwr_t::pending_writes(), buf_dblwr_t::writes_pending:
Keeps track of writes of persistent data pages.
buf_flush_LRU(): Allow calls while LRU flushing may be in progress
in another thread.
Tested by Matthias Leich (correctness) and Axel Schwenke (performance)
Adaptive flushing is enabled by setting innodb_max_dirty_pages_pct_lwm>0
(not default) and innodb_adaptive_flushing=ON (default).
There is also the parameter innodb_adaptive_flushing_lwm
(default: 10 per cent of the log capacity). It should enable some
adaptive flushing even when innodb_max_dirty_pages_pct_lwm=0.
That is not being changed here.
This idea was first presented by Inaam Rana several years ago,
and I discussed it with Jean-François Gagné at FOSDEM 2023.
buf_flush_page_cleaner(): When we are not near the log capacity limit
(neither buf_flush_async_lsn nor buf_flush_sync_lsn are set),
also try to move clean blocks from the buf_pool.LRU list to buf_pool.free
or initiate writes (but not the eviction) of dirty blocks, until
the remaining I/O capacity has been consumed.
buf_flush_LRU_list_batch(): Add the parameter bool evict, to specify
whether dirty least recently used pages (from buf_pool.LRU) should
be evicted immediately after they have been written out. Callers outside
buf_flush_page_cleaner() will pass evict=true, to retain the existing
behaviour.
buf_do_LRU_batch(): Add the parameter bool evict.
Return counts of evicted and flushed pages.
buf_flush_LRU(): Add the parameter bool evict.
Assume that the caller holds buf_pool.mutex and
will invoke buf_dblwr.flush_buffered_writes() afterwards.
buf_flush_list_holding_mutex(): A low-level variant of buf_flush_list()
whose caller must hold buf_pool.mutex and invoke
buf_dblwr.flush_buffered_writes() afterwards.
buf_flush_wait_batch_end_acquiring_mutex(): Remove. It is enough to have
buf_flush_wait_batch_end().
page_cleaner_flush_pages_recommendation(): Avoid some floating-point
arithmetics.
buf_flush_page(), buf_flush_check_neighbor(), buf_flush_check_neighbors(),
buf_flush_try_neighbors(): Rename the parameter "bool lru" to "bool evict".
buf_free_from_unzip_LRU_list_batch(): Remove the parameter.
Only actual page writes will contribute towards the limit.
buf_LRU_free_page(): Evict freed pages of temporary tables.
buf_pool.done_free: Broadcast whenever a block is freed
(and buf_pool.try_LRU_scan is set).
buf_pool_t::io_buf_t::reserve(): Retry indefinitely.
During the test encryption.innochecksum we easily run out of
these buffers for PAGE_COMPRESSED or ENCRYPTED pages.
Tested by Matthias Leich and Axel Schwenke
- During non-last batch of multi-batch recovery, InnoDB holds
log_sys.mutex and preallocates the block which may intiate
page flush, which may initiate log flush, which requires
log_sys.mutex to acquire again. This leads to assert failure.
So InnoDB recovery should release log_sys.mutex before
preallocating the block.
In commit a03dd94be8 as well as
mysql/mysql-server@6ef8c34344
the iterations were changed so that the variable "scanned"
would remain 0 when the first list item qualifies for eviction.
buf_LRU_free_from_unzip_LRU_list(), buf_LRU_free_from_common_LRU_list():
Increment "scanned" when a block can be freed.
buf_LRU_free_from_common_LRU_list(): Remove a redundant condition.
Whenever this function is invoked, buf_pool.LRU should be nonempty,
hence something should always be scanned.
Thanks to Jean-François Gagné for reporting this.
- InnoDB AHI tries to access the concurrent instant alter column,
leads to asan failure. Instant alter column should acquire the
clustered index search latch in exclusive mode before changing
the table cache definition.
- Removed the default parameter for the function
btr_search_drop_page_hash_index()
- Addressed the DWITH_INNODB_AHI=0 compilation failure
by passing two parameters from all callers of
btr_search_drop_page_hash_index()
Since commit 0b47c126e3 (MDEV-13542)
we treat all-zero pages as corrupted ones.
During a stress test, a read-ahead of an all-zero page was triggered
and the page read was completed concurrently with buf_page_create_low().
This caused the assertion to fail, because buf_page_create_low() was
waiting for the page latch.
buf_page_get_low(): Only invoke buf_pool_t::corrupted_evict()
if the block was not already marked as corrupted.
buf_page_create_low(): On page identifier mismatch, retry the
buf_pool.page_hash lookup.
buf_pool_t::corrupted_evict(): Set the state of the block to FREED
so that a concurrent buf_page_get_low() will refuse to load the page.
Wait for the page latch to be vacant before proceeding to remove
the block from buf_pool.page_hash and buf_pool.LRU.
page_id_t::set_corrupted(), page_id_t::is_corrupted(): Accessors
for indicating a corrupted page identifier.
Tested by Matthias Leich
The approach to handling corruption that was chosen by Oracle in
commit 177d8b0c12
is not really useful. Not only did it actually fail to prevent InnoDB
from crashing, but it is making things worse by blocking attempts to
rescue data from or rebuild a partially readable table.
We will try to prevent crashes in a different way: by propagating
errors up the call stack. We will never mark the clustered index
persistently corrupted, so that data recovery may be attempted by
reading from the table, or by rebuilding the table.
This should also fix MDEV-13680 (crash on btr_page_alloc() failure);
it was extensively tested with innodb_file_per_table=0 and a
non-autoextend system tablespace.
We should now avoid crashes in many cases, such as when a page
cannot be read or allocated, or an inconsistency is detected when
attempting to update multiple pages. We will not crash on double-free,
such as on the recovery of DDL in system tablespace in case something
was corrupted.
Crashes on corrupted data are still possible. The fault injection mechanism
that is introduced in the subsequent commit may help catch more of them.
buf_page_import_corrupt_failure: Remove the fault injection, and instead
corrupt some pages using Perl code in the tests.
btr_cur_pessimistic_insert(): Always reserve extents (except for the
change buffer), in order to prevent a subsequent allocation failure.
btr_pcur_open_at_rnd_pos(): Merged to the only caller ibuf_merge_pages().
btr_assert_not_corrupted(), btr_corruption_report(): Remove.
Similar checks are already part of btr_block_get().
FSEG_MAGIC_N_BYTES: Replaces FSEG_MAGIC_N_VALUE.
dict_hdr_get(), trx_rsegf_get_new(), trx_undo_page_get(),
trx_undo_page_get_s_latched(): Replaced with error-checking calls.
trx_rseg_t::get(mtr_t*): Replaces trx_rsegf_get().
trx_rseg_header_create(): Let the caller update the TRX_SYS page if needed.
trx_sys_create_sys_pages(): Merged with trx_sysf_create().
dict_check_tablespaces_and_store_max_id(): Do not access
DICT_HDR_MAX_SPACE_ID, because it was already recovered in dict_boot().
Merge dict_check_sys_tables() with this function.
dir_pathname(): Replaces os_file_make_new_pathname().
row_undo_ins_remove_sec(): Do not modify the undo page by adding
a terminating NUL byte to the record.
btr_decryption_failed(): Report decryption failures
dict_set_corrupted_by_space(), dict_set_encrypted_by_space(),
dict_set_corrupted_index_cache_only(): Remove.
dict_set_corrupted(): Remove the constant parameter dict_locked=false.
Never flag the clustered index corrupted in SYS_INDEXES, because
that would deny further access to the table. It might be possible to
repair the table by executing ALTER TABLE or OPTIMIZE TABLE, in case
no B-tree leaf page is corrupted.
dict_table_skip_corrupt_index(), dict_table_next_uncorrupted_index(),
row_purge_skip_uncommitted_virtual_index(): Remove, and refactor
the callers to read dict_index_t::type only once.
dict_table_is_corrupted(): Remove.
dict_index_t::is_btree(): Determine if the index is a valid B-tree.
BUF_GET_NO_LATCH, BUF_EVICT_IF_IN_POOL: Remove.
UNIV_BTR_DEBUG: Remove. Any inconsistency will no longer trigger
assertion failures, but error codes being returned.
buf_corrupt_page_release(): Replaced with a direct call to
buf_pool.corrupted_evict().
fil_invalid_page_access_msg(): Never crash on an invalid read;
let the caller of buf_page_get_gen() decide.
btr_pcur_t::restore_position(): Propagate failure status to the caller
by returning CORRUPTED.
opt_search_plan_for_table(): Simplify the code.
row_purge_del_mark(), row_purge_upd_exist_or_extern_func(),
row_undo_ins_remove_sec_rec(), row_undo_mod_upd_del_sec(),
row_undo_mod_del_mark_sec(): Avoid mem_heap_create()/mem_heap_free()
when no secondary indexes exist.
row_undo_mod_upd_exist_sec(): Simplify the code.
row_upd_clust_step(), dict_load_table_one(): Return DB_TABLE_CORRUPT
if the clustered index (and therefore the table) is corrupted, similar
to what we do in row_insert_for_mysql().
fut_get_ptr(): Replace with buf_page_get_gen() calls.
buf_page_get_gen(): Return nullptr and *err=DB_CORRUPTION
if the page is marked as freed. For other modes than
BUF_GET_POSSIBLY_FREED or BUF_PEEK_IF_IN_POOL this will
trigger a debug assertion failure. For BUF_GET_POSSIBLY_FREED,
we will return nullptr for freed pages, so that the callers
can be simplified. The purge of transaction history will be
a new user of BUF_GET_POSSIBLY_FREED, to avoid crashes on
corrupted data.
buf_page_get_low(): Never crash on a corrupted page, but simply
return nullptr.
fseg_page_is_allocated(): Replaces fseg_page_is_free().
fts_drop_common_tables(): Return an error if the transaction
was rolled back.
fil_space_t::set_corrupted(): Report a tablespace as corrupted if
it was not reported already.
fil_space_t::io(): Invoke fil_space_t::set_corrupted() to report
out-of-bounds page access or other errors.
Clean up mtr_t::page_lock()
buf_page_get_low(): Validate the page identifier (to check for
recently read corrupted pages) after acquiring the page latch.
buf_page_t::read_complete(): Flag uninitialized (all-zero) pages
with DB_FAIL. Return DB_PAGE_CORRUPTED on page number mismatch.
mtr_t::defer_drop_ahi(): Renamed from mtr_defer_drop_ahi().
recv_sys_t::free_corrupted_page(): Only set_corrupt_fs()
if any log records exist for the page. We do not mind if read-ahead
produces corrupted (or all-zero) pages that were not actually needed
during recovery.
recv_recover_page(): Return whether the operation succeeded.
recv_sys_t::recover_low(): Simplify the logic. Check for recovery error.
Thanks to Matthias Leich for testing this extensively and to the
authors of https://rr-project.org for making it easy to diagnose
and fix any failures that were found during the testing.
During an increase in resize, the new curr_size got a value
less than old_size.
As n_chunks_new and n_chunks have a strong correlation to the
resizing operation in progress, we can use them and remove the
need for old_size.
For convienece the n_chunks_new < n_chunks is now the is_shrinking
function.
The volatile compiler optimization on n_chunks{,_new} is removed
as real mutex uses are needed.
Other n_chunks_new/n_chunks methods:
n_chunks_new and n_chunks almost always read and altered under
the pool mutex.
Exceptions are:
* i_s_innodb_buffer_page_fill,
* buf_pool_t::is_uncompressed (via is_blocked_field)
These need reexamining for the need of a mutex, however comments
indicates this already.
get_n_pages has uses in buffer pool load, recover log memory
exhaustion estimates and innodb status so take the minimum number
of chunks for safety.
The buf_pool_t::running_out function also uses curr_size/old_size.
We replace this hot function calculation with just n_chunks_new.
This is the new size of the chunks before the resizing occurs.
If we are resizing down, we've already got the case we had previously
(as the minimum). If we are resizing upwards, we are taking an
optimistic view that there will be buffer chunks available for locks.
As this memory allocation is occurring immediately next the resizing
function it seems likely.
Compiler hint UNIV_UNLIKELY removed to leave it to the branch
predictor to make an informed decision.
Added test case of a smaller size than the Marko/Roel original
in JIRA reducing the size to 256M. SEGV hits roughly 1/10 times
but its better than a 21G memory size.
Reviewer: Marko
buf_LRU_scan_and_free_block(): It turns out that even with
-fno-expensive-optimizations, GCC 4.8.5 may fail to split an instruction.
For the non-embedded server, -O1 would fail and -Og would seem to work,
while the embedded server build seems to require -O0.
buf_block_init(): Correct the MemorySanitizer instrumentation.
buf_page_get_low(): Do not read dirty data from read-fixed blocks.
These data races were identified by MemorySanitizer. If a read-fixed
block is being accessed, we must acquire and release a page latch,
so that the read-fix (and the exclusive page latch) will be released
and it will be safe to read the page frame contents if needed,
even before acquiring the final page latch. We do that in
buf_read_ahead_linear() and for the allow_ibuf_merge check.
mtr_t::page_lock(): Assert that the block is not read-fixed.
buf_page_t::frame: Moved from buf_block_t::frame.
All 'thin' buf_page_t describing compressed-only ROW_FORMAT=COMPRESSED
pages will have frame=nullptr, while all 'fat' buf_block_t
will have a non-null frame pointing to aligned innodb_page_size bytes.
This eliminates the need for separate states for
BUF_BLOCK_FILE_PAGE and BUF_BLOCK_ZIP_PAGE.
buf_page_t:🔒 Moved from buf_block_t::lock. That is, all block
descriptors will have a page latch. The IO_PIN state that was used
for discarding or creating the uncompressed page frame of a
ROW_FORMAT=COMPRESSED block is replaced by a combination of read-fix
and page X-latch.
page_zip_des_t::fix: Replaces state_, buf_fix_count_, io_fix_, status
of buf_page_t with a single std::atomic<uint32_t>. All modifications
will use store(), fetch_add(), fetch_sub(). This space was previously
wasted to alignment on 64-bit systems. We will use the following encoding
that combines a state (partly read-fix or write-fix) and a buffer-fix
count:
buf_page_t::NOT_USED=0 (previously BUF_BLOCK_NOT_USED)
buf_page_t::MEMORY=1 (previously BUF_BLOCK_MEMORY)
buf_page_t::REMOVE_HASH=2 (previously BUF_BLOCK_REMOVE_HASH)
buf_page_t::FREED=3 + fix: pages marked as freed in the file
buf_page_t::UNFIXED=1U<<29 + fix: normal pages
buf_page_t::IBUF_EXIST=2U<<29 + fix: normal pages; may need ibuf merge
buf_page_t::REINIT=3U<<29 + fix: reinitialized pages (skip doublewrite)
buf_page_t::READ_FIX=4U<<29 + fix: read-fixed pages (also X-latched)
buf_page_t::WRITE_FIX=5U<<29 + fix: write-fixed pages (also U-latched)
buf_page_t::WRITE_FIX_IBUF=6U<<29 + fix: write-fixed; may have ibuf
buf_page_t::WRITE_FIX_REINIT=7U<<29 + fix: write-fixed (no doublewrite)
buf_page_t::write_complete(): Change WRITE_FIX or WRITE_FIX_REINIT to
UNFIXED, and WRITE_FIX_IBUF to IBUF_EXIST, before releasing the U-latch.
buf_page_t::read_complete(): Renamed from buf_page_read_complete().
Change READ_FIX to UNFIXED or IBUF_EXIST, before releasing the X-latch.
buf_page_t::can_relocate(): If the page latch is being held or waited for,
or the block is buffer-fixed or io-fixed, return false. (The condition
on the page latch is new.)
Outside buf_page_get_gen(), buf_page_get_low() and buf_page_free(), we
will acquire the page latch before fix(), and unfix() before unlocking.
buf_page_t::flush(): Replaces buf_flush_page(). Optimize the
handling of FREED pages.
buf_pool_t::release_freed_page(): Assume that buf_pool.mutex is held
by the caller.
buf_page_t::is_read_fixed(), buf_page_t::is_write_fixed(): New predicates.
buf_page_get_low(): Ignore guesses that are read-fixed because they
may not yet be registered in buf_pool.page_hash and buf_pool.LRU.
buf_page_optimistic_get(): Acquire latch before buffer-fixing.
buf_page_make_young(): Leave read-fixed blocks alone, because they
might not be registered in buf_pool.LRU yet.
recv_sys_t::recover_deferred(), recv_sys_t::recover_low():
Possibly fix MDEV-26326, by holding a page X-latch instead of
only buffer-fixing the page.
In commit c091a0bc8d we removed
the use of the HASH_ macros for inserting into
buf_pool.page_hash, or accessing buf_page_t::hash.
However, the binary buddy allocator for block->page.zip.data would
still use the HASH_ macros. HASH_INSERT and not HASH_DELETE would reset
the next-block pointer to the null pointer. Our replacement of
HASH_DELETE() will reset the next-block pointer, and the replacement of
HASH_INSERT() assumes that the pointer is the null pointer.
buf_LRU_block_free_non_file_page(): Assert that the next-block pointer
is the null pointer.
buf_buddy_block_free(): Reset the pointer before invoking
buf_LRU_block_free_non_file_page(). Without this, the added
assertion would fail in the test encryption.innochecksum.
This implements memory transaction support for:
* Intel Restricted Transactional Memory (RTM), also known as TSX-NI
(Transactional Synchronization Extensions New Instructions)
* POWER v2.09 Hardware Trace Monitor (HTM) on GNU/Linux
transactional_lock_guard, transactional_shared_lock_guard:
RAII lock guards that try to elide the lock acquisition
when transactional memory is available.
buf_pool.page_hash: Try to elide latches whenever feasible.
Related to the InnoDB change buffer and ROW_FORMAT=COMPRESSED
tables, this is not always possible.
In buf_page_get_low(), memory transactions only work reasonably
well for validating a guessed block address.
TMLockGuard, TMLockTrxGuard, TMLockMutexGuard: RAII lock guards
that try to elide lock_sys.latch and related latches.
Since commit bd5a6403ca (MDEV-26033)
we can actually calculate the buf_pool.page_hash cell and latch
addresses while not holding buf_pool.mutex.
buf_page_alloc_descriptor(): Remove the MEM_UNDEFINED.
We now expect buf_page_t::hash to be zero-initialized.
buf_pool_t::hash_chain: Dedicated data type for buf_pool.page_hash.array.
buf_LRU_free_one_page(): Merged to the only caller
buf_pool_t::corrupted_evict().
The lazy deletion of clean blocks from buf_pool.flush_list that was
introduced in commit 6441bc614a (MDEV-25113)
introduced a race condition around the function
buf_flush_relocate_on_flush_list().
The test innodb_zip.wl5522_debug_zip as well as the buffer pool
resizing tests would occasionally fail in debug builds due to
buf_pool.flush_list.count disagreeing with the actual length of the
doubly-linked list.
The safe procedure for relocating a block in buf_pool.flush_list should be
as follows, now that we implement lazy deletion from buf_pool.flush_list:
1. Acquire buf_pool.mutex.
2. Acquire the exclusive buf_pool.page_hash.latch.
3. Acquire buf_pool.flush_list_mutex.
4. Copy the block descriptor.
5. Invoke buf_flush_relocate_on_flush_list().
6. Release buf_pool.flush_list_mutex.
buf_flush_relocate_on_flush_list(): Assert that
buf_pool.flush_list_mutex is being held. Invoke
buf_page_t::oldest_modification() only once, using
std::memory_order_relaxed, now that the mutex protects us.
buf_LRU_free_page(), buf_LRU_block_remove_hashed(): Avoid
an unlock-lock cycle on hash_lock. (We must not acquire hash_lock
while already holding buf_pool.flush_list_mutex, because that
could lead to a deadlock due to latching order violation.)
In commit 5f22511e35 we depend on
Total Store Ordering. For correct operation on ISAs that implement
weaker memory ordering, we must explicitly use release/acquire stores
and loads on buf_page_t::oldest_modification_ to prevent a race condition
when buf_page_t::list does not happen to be on the same cache line.
buf_page_t::clear_oldest_modification(): Assert that the block is
not in buf_pool.flush_list, and use std::memory_order_release.
buf_page_t::oldest_modification_acquire(): Read oldest_modification_
with std::memory_order_acquire. In this way, if the return value is 0,
the caller may safely assume that it will not observe the buf_page_t
as being in buf_pool.flush_list, even if it is not holding
buf_pool.flush_list_mutex.
buf_flush_relocate_on_flush_list(), buf_LRU_free_page():
Invoke buf_page_t::oldest_modification_acquire().
buf_LRU_get_free_block(): Initially wait for a single block to be
freed, signaled by buf_pool.done_free. Only if that fails and no
LRU eviction flushing batch is already running, we initiate a
flushing batch that should serve all threads that are currently
waiting in buf_LRU_get_free_block().
Note: In an extreme case, this may introduce a performance regression
at larger numbers of connections. We observed this in sysbench
oltp_update_index with 512MiB buffer pool, 4GiB of data on fast NVMe,
and 1000 concurrent connections, on a 20-thread CPU. The contention point
appears to be buf_pool.mutex, and the improvement would turn into a
regression somewhere beyond 32 concurrent connections.
On slower storage, such regression was not observed; instead, the
throughput was improving and maximum latency was reduced.
The excessive waits were pointed out by Vladislav Vaintroub.
buf_page_write_complete(): Reduce the buf_pool.mutex hold time,
and do not acquire buf_pool.flush_list_mutex at all.
Instead, mark blocks clean by setting oldest_modification to 1.
Dirty pages of temporary tables will be identified by the special
value 2 instead of the previous special value 1.
(By design of the ib_logfile0 format, actual LSN values smaller
than 2048 are not possible.)
buf_LRU_free_page(), buf_pool_t::get_oldest_modification()
and many other functions will remove the garbage (clean blocks)
from buf_pool.flush_list while holding buf_pool.flush_list_mutex.
buf_pool_t::n_flush_LRU, buf_pool_t::n_flush_list:
Replaced with non-atomic variables, protected by buf_pool.mutex,
to avoid unnecessary synchronization when modifying the counts.
export_vars: Remove unnecessary indirection for
innodb_pages_created, innodb_pages_read, innodb_pages_written.
In commit 7cffb5f6e8 (MDEV-23399)
the implementation of buf_flush_dirty_pages() was replaced with
a slow one, which would perform excessive scans of the
buf_pool.flush_list and make little progress.
buf_flush_list(), buf_flush_LRU(): Split from buf_flush_lists().
Vladislav Vaintroub noticed that we will not need to invoke
log_flush_task.wait() for the LRU eviction flushing.
buf_flush_list_space(): Replaces buf_flush_dirty_pages().
This is like buf_flush_list(), but operating on a single
tablespace at a time. Writes at most innodb_io_capacity
pages. Returns whether some of the tablespace might remain
in the buffer pool.
buf_LRU_get_free_block(): Increment the counter after acquiring
buf_pool.mutex.
buf_pool.stat.LRU_waits: Replaces export_vars.innodb_buffer_pool_wait_free
and srv_stats.buf_pool_wait_free.
Reads of the counter will remain dirty, not acquiring the mutex.
SHOW ENGINE INNODB MUTEX functionality is completely removed,
as are the InnoDB latching order checks.
We will enforce innodb_fatal_semaphore_wait_threshold
only for dict_sys.mutex and lock_sys.mutex.
dict_sys_t::mutex_lock(): A single entry point for dict_sys.mutex.
lock_sys_t::mutex_lock(): A single entry point for lock_sys.mutex.
FIXME: srv_sys should be removed altogether; it is duplicating tpool
functionality.
fil_crypt_threads_init(): To prevent SAFE_MUTEX warnings, we must
not hold fil_system.mutex.
fil_close_all_files(): To prevent SAFE_MUTEX warnings for
fil_space_destroy_crypt_data(), we must not hold fil_system.mutex
while invoking fil_space_free_low() on a detached tablespace.
InnoDB buffer pool block and index tree latches depend on a
special kind of read-update-write lock that allows reentrant
(recursive) acquisition of the 'update' and 'write' locks
as well as an upgrade from 'update' lock to 'write' lock.
The 'update' lock allows any number of reader locks from
other threads, but no concurrent 'update' or 'write' lock.
If there were no requirement to support an upgrade from 'update'
to 'write', we could compose the lock out of two srw_lock
(implemented as any type of native rw-lock, such as SRWLOCK on
Microsoft Windows). Removing this requirement is very difficult,
so in commit f7e7f487d4b06695f91f6fbeb0396b9d87fc7bbf we
implemented an 'update' mode to our srw_lock.
Re-entrant or recursive locking is mostly needed when writing or
freeing BLOB pages, but also in crash recovery or when merging
buffered changes to an index page. The re-entrancy allows us to
attach a previously acquired page to a sub-mini-transaction that
will be committed before whatever else is holding the page latch.
The SUX lock supports Shared ('read'), Update, and eXclusive ('write')
locking modes. The S latches are not re-entrant, but a single S latch
may be acquired even if the thread already holds an U latch.
The idea of the U latch is to allow a write of something that concurrent
readers do not care about (such as the contents of BTR_SEG_LEAF,
BTR_SEG_TOP and other page allocation metadata structures, or
the MDEV-6076 PAGE_ROOT_AUTO_INC). (The PAGE_ROOT_AUTO_INC field
is only updated when a dict_table_t for the table exists, and only
read when a dict_table_t for the table is being added to dict_sys.)
block_lock::u_lock_try(bool for_io=true) is used in buf_flush_page()
to allow concurrent readers but no concurrent modifications while the
page is being written to the data file. That latch will be released
by buf_page_write_complete() in a different thread. Hence, we use
the special lock owner value FOR_IO.
The index_lock::u_lock() improves concurrency on operations that
involve non-leaf index pages.
The interface has been cleaned up a little. We will use
x_lock_recursive() instead of x_lock() when we know that a
lock is already held by the current thread. Similarly,
a lock upgrade from U to X is only allowed via u_x_upgrade()
or x_lock_upgraded() but not via x_lock().
We will disable the LatchDebug and sync_array interfaces to
InnoDB rw-locks.
The SEMAPHORES section of SHOW ENGINE INNODB STATUS output
will no longer include any information about InnoDB rw-locks,
only TTASEventMutex (cmake -DMUTEXTYPE=event) waits.
This will make a part of the 'innotop' script dead code.
The block_lock buf_block_t::lock will not be covered by any
PERFORMANCE_SCHEMA instrumentation.
SHOW ENGINE INNODB MUTEX and INFORMATION_SCHEMA.INNODB_MUTEXES
will no longer output source code file names or line numbers.
The dict_index_t::lock will be identified by index and table names,
which should be much more useful. PERFORMANCE_SCHEMA is lumping
information about all dict_index_t::lock together as
event_name='wait/synch/sxlock/innodb/index_tree_rw_lock'.
buf_page_free(): Remove the file,line parameters. The sux_lock will
not store such diagnostic information.
buf_block_dbg_add_level(): Define as empty macro, to be removed
in a subsequent commit.
Unless the build was configured with cmake -DPLUGIN_PERFSCHEMA=NO
the index_lock dict_index_t::lock will be instrumented via
PERFORMANCE_SCHEMA. Similar to
commit 1669c8890c
we will distinguish lock waits by registering shared_lock,exclusive_lock
events instead of try_shared_lock,try_exclusive_lock.
Actual 'try' operations will not be instrumented at all.
rw_lock_list: Remove. After MDEV-24167, this only covered
buf_block_t::lock and dict_index_t::lock. We will output their
information by traversing buf_pool or dict_sys.
A side effect of MDEV-16264 is that a large number of threads will
be created at server startup, to be destroyed after a minute or two.
One source of such thread creation is srv_start_periodic_timer().
InnoDB is creating 3 periodic tasks: srv_master_callback (1Hz)
srv_error_monitor_task (1Hz), and srv_monitor_task (0.2Hz).
It appears that we can merge srv_error_monitor_task and srv_monitor_task
and have them invoked 4 times per minute (every 15 seconds). This will
affect our ability to enforce innodb_fatal_semaphore_wait_threshold and
some computations around BUF_LRU_STAT_N_INTERVAL.
We could remove srv_master_callback along with the DROP TABLE queue
at some point of time in the future. We must keep it independent
of the innodb_fatal_semaphore_wait_threshold detection, because
the background DROP TABLE queue could get stuck due to dict_sys
being locked by another thread. For now, srv_master_callback
must be invoked once per second, so that
innodb_flush_log_at_timeout=1 can work.
BUF_LRU_STAT_N_INTERVAL: Reduce the precision and extend the time
from 50*1 second to 4*15 seconds.
srv_error_monitor_timer: Remove.
MAX_MUTEX_NOWAIT: Increase from 20*1 second to 2*15 seconds.
srv_refresh_innodb_monitor_stats(): Avoid a repeated call to time(NULL).
Change the interval to less than 60 seconds.
srv_monitor(): Renamed from srv_monitor_task.
srv_monitor_task(): Renamed from srv_error_monitor_task().
Invoked only once in 15 seconds. Invoke also srv_monitor().
Increase the fatal_cnt threshold from 10*1 second to 1*15 seconds.
sync_array_print_long_waits_low(): Invoke time(NULL) only once.
Remove a bogus message about printouts for 30 seconds. Those
printouts were effectively already disabled in MDEV-16264
(commit 5e62b6a5e0).
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 d1ab89037a 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
An unfortunate performance regression was introduced in
commit b1ab211dee (MDEV-15053)
that affects workloads where the working set is larger than
the buffer pool.
We used to terminate the loop in buf_LRU_scan_and_free_block()
when we succeeded to free a replaceable buffer pool block.
Due to the regression, we would keep freeing all replaceable
blocks until we reach the end of the list or encounter a block
that had never been accessed because it had been brought in
by read-ahead.
The srv_monitor_event and the srv_monitor_thread would not be
created when InnoDB is in read-only mode. Yet, some code would
unconditionally invoke os_event_set(srv_monitor_event).
In commit 0f90728bc0 (MDEV-16809)
we introduced the configuration option innodb_log_optimize_ddl
for controlling whether native index creation or table-rebuild
in InnoDB should avoid writing full redo log.
Fungo Wang reported that this option is causing occasional failures.
The reason is that pages may be written to data files in an
inconsistent state. Applying log records to such inconsistent pages
may fail.
The solution is to always invoke PageBulk::finish() before page latches
may be released, to ensure that the page contents is in a consistent
state.
Something similar was implemented in MySQL 8.0.13:
mysql/mysql-server@d1254b9473
buf_block_t::skip_flush_check: Remove. Suppressing consistency checks
is a bad idea.
PageBulk::needs_finish(): New predicate: Determine whether
PageBulk::finish() must fix up the page.
PageBulk::init(): Clear PAGE_DIRECTION to ensure that needs_finish()
will hold. We change the field from PAGE_NO_DIRECTION to 0
and back without writing redo log. This trick avoids the need
to introduce any new data member to PageBulk.
PageBulk::insert(): Replace some high-level accessors to bypass
debug assertions related to PAGE_HEAP_TOP that we will be violating
until finish() has been executed.
PageBulk::finish(): Tolerate m_rec_no==0. We must invoke this also
on an empty page, to ensure that PAGE_HEAP_TOP is initialized.
PageBulk::commit(): Always invoke finish().
PageBulk::release(), BtrBulk::pageSplit(), BtrBulk::storeExt(),
BtrBulk::finish(): Invoke PageBulk::finish().
