Issue: During mtr_t:commit, if there is not enough space available in
redo log buffer, we flush the buffer. During flush, the LSN lock is
released allowing other concurrent mtr to commit. After flush we
reacquire the lock but use the old LSN obtained before check. It could
lead to redo log corruption. As the LSN moves backwards with the
possibility of data loss and unrecoverable server if the server aborts
for any reason or if server is shutdown with innodb_fast_shutdown=2.
With normal shutdown, recovery fails to map the checkpoint LSN to
correct offset.
In debug mode it hits log0log.cc:863: lsn_t log_t::write_buf()
Assertion `new_buf_free == ((lsn - first_lsn) & write_size_1)' failed.
In release mode, after normal shutdown, restart fails.
[ERROR] InnoDB: Missing FILE_CHECKPOINT(8416546) at 8416546
[ERROR] InnoDB: Log scan aborted at LSN 8416546
Backup fails reading the corrupt redo log.
[00] 2024-07-31 20:59:10 Retrying read of log at LSN=7334851
[00] FATAL ERROR: 2024-07-31 20:59:11 Was only able to copy log from
7334851 to 7334851, not 8416446; try increasing innodb_log_file_size
Unless a backup is tried or the server is shutdown or killed
immediately, the corrupt redo part is eventually truncated and there
may not be any visible issues seen in release mode.
This issue was introduced by the following commit.
commit a635c40648
MDEV-27774 Reduce scalability bottlenecks in mtr_t::commit()
Fix: If we need to release latch and flush redo before writing mtr
logs, make sure to get the latest system LSN after reacquiring the
redo system latch.
mtr_t::commit_shrink(): Do not assert that some previously clean pages
will be flagged as modified by this mini-transaction. It could be the
case that there had been no recent write-back of any of the undo
tablespace pages that we are modifying when truncating the tablespace.
It suffices to assert that some pages were modified again:
ut_ad(m_modifications).
This fixes up commit f5fddae3cb
In so-called optimistic buffer pool lookups, we must not
dereference a block descriptor before we have made sure that
it is accessible. While buf_pool_t::resize() is running,
block descriptors could become invalid.
The buf::Block_hint class was essentially duplicating
a buf_pool.page_hash lookup that was executed in
buf_page_optimistic_get() anyway. For better locality of
reference, we had better execute that lookup only once.
buf_page_optimistic_fix(): Prepare for buf_page_optimistic_get().
This basically is a simpler version of Buf::Block_hint.
buf_page_optimistic_get(): Assume that buf_page_optimistic_fix()
has been called and the page identifier verified. Should the block
be evicted, the block->modify_clock will be invalidated; we do not
need to check the block->page.id() again. It suffices to check
the block->modify_clock after acquiring the page latch.
btr_pcur_t::old_page_id: Store the expected page identifier
for buf_page_optimistic_fix().
btr_pcur_t::block_when_stored: Remove. This was duplicating
page_cur_t::block.
btr_pcur_optimistic_latch_leaves(): Remove redundant parameters.
First, invoke buf_page_optimistic_fix() on the requested page.
If needed, acquire a latch on the left page. Finally, acquire
a latch on the target page and recheck the block->modify_clock.
If the page had been freed while we were not holding a page latch,
fall back to the slow path. Validate the FIL_PAGE_PREV after
acquiring a latch on the current page. The block->modify_clock
is only being incremented when records are deleted or pages
reorganized or evicted; it does not guard against concurrent
page splits.
Reviewed by: Debarun Banerjee
log_t: Define buf_size, max_buf_free as 32-bit and next_checkpoint_no
as byte (we only need a bit) and rearrange some data members,
so that on AMD64 we can fit log_sys.latch and log_sys.log in
the same 64-byte cache line.
mtr_t::commit_log(), mtr_t::commit_logger: A part of mtr_t::commit()
split into a separate function, so that we will not unnecessarily invoke
log_sys.get_write_target() when running on a memory-mapped log file,
or log_sys.is_pmem().
Reviewed by: Vladislav Vaintroub
Tested by: Matthias Leich
Some fixes related to commit f838b2d799 and
Rows_log_event::do_apply_event() and Update_rows_log_event::do_exec_row()
for system-versioned tables were provided by Nikita Malyavin.
This was required by test versioning.rpl,trx_id,row.
innodb_log_spin_wait_delay_update(): Always acquire log_sys.latch
to protect the change of mtr_t::spin_wait_delay.
log_t::lock_lsn(): In the general case, actually use
mtr_t::spin_wait_delay as it was intended. In the x86 specific
log_t::lock_lsn_bts() we used mtr_t::spin_wait_delay.
The log_sys.lsn_lock is a very contended resource with a small
critical section in log_sys.append_prepare(). On many processor
microarchitectures, replacing the system call based log_sys.lsn_lock
with a pure spin lock would fare worse during high concurrency workloads,
wasting a significant amount of CPU cycles in the spin loop.
On other microarchitectures, we would see a significant amount of time
being spent in native_queued_spin_lock_slowpath() in the Linux kernel,
plus context switching between user and kernel address space. This was
pointed out by Steve Shaw from Intel Corporation.
Depending on the workload and the hardware implementation, it may be
useful to use a pure spin lock in log_sys.append_prepare().
We will introduce a parameter. The statement
SET GLOBAL INNODB_LOG_SPIN_WAIT_DELAY=50;
would enable a spin lock that will execute that many MY_RELAX_CPU()
operations (such as the x86 PAUSE instruction) between successive
attempts of acquiring the spin lock. The use of a system call based
log_sys.lsn_lock (which is the default setting) can be enabled by
SET GLOBAL INNODB_LOG_SPIN_WAIT_DELAY=0;
This patch will also introduce #ifdef LOG_LATCH_DEBUG
(part of cmake -DWITH_INNODB_EXTRA_DEBUG=ON) for more accurate
tracking of log_sys.latch ownership and reorganize the fields of
log_sys to improve the locality of reference and to reduce the
chances of false sharing.
When a spin lock is being used, it will be maintained in the
most significant bit of log_sys.buf_free. This is useful, because that is
one of the fields that is covered by the lock. For IA-32 or AMD64, we
implement the spin lock specially via log_t::lsn_lock_bts(), employing the
i386 LOCK BTS instruction. A straightforward std::atomic::fetch_or() would
translate into an inefficient loop around LOCK CMPXCHG.
mtr_t::spin_wait_delay: The value of innodb_log_spin_wait_delay.
mtr_t::finisher: Pointer to the currently used mtr_t::finish_write()
implementation. This allows to avoid introducing conditional branches.
We no longer invoke log_sys.is_pmem() at the mini-transaction level,
but we would do that in log_write_up_to().
mtr_t::finisher_update(): Update finisher when spin_wait_delay is
changed from or to 0 (the spin lock is changed to log_sys.lsn_lock or
vice versa).
Problem:
========
During upgrade, InnoDB does write the redo log for adjusting
the tablespace size or tablespace flags even before the log
has upgraded to configured format. This could lead to data
inconsistent if any crash happened during upgrade process.
Fix:
===
srv_start(): Write the tablespace flags adjustment, increased
tablespace size redo log only after redo log upgradation.
log_write_low(), log_reserve_and_write_fast(): Check whether
the redo log is in physical format.
This regression is introduced in MDEV-28708 where the MTR_LOG_NO_REDO
mtrs are assigned last_checkpoint_lsn as the LSN. It causes a race with
checkpoint in pending state. The concurrent checkpoint writes a
checkpoint LSN of larger value after pages with older checkpoint LSN is
inserted into the flush list. The next checkpoint sees the reversal in
checkpoint sequence and asserts if the pages are not yet flushed.
There could be several ways to solve this issue. Ideally the unlogged
mtr should take the latest LSN as opposed to going behind and use the
previous checkpoint LSN. It has been the older design and seems good.
Also, other than the critical race, using the old checkpoint LSN adds
the pages to other end of flush list overriding all existing dirty
pages and looks counter intuitive.
While the index_lock and block_lock include debug instrumentation
to keep track of shared lock holders, such instrumentation was
never part of the simpler srw_lock, and therefore some users of the
class implemented a limited form of bookkeeping.
srw_lock_debug encapsulates srw_lock and adds the data members
writer, readers_lock, and readers to keep track of the threads that
hold the exclusive latch or any shared latches. The debug checks
are available also with SUX_LOCK_GENERIC (in environments that do not
implement a futex-like system call).
dict_sys_t::latch: Use srw_lock_debug in debug builds.
This makes the debug fields latch_ex, latch_readers redundant.
fil_space_t::latch: Use srw_lock_debug in debug builds.
This makes the debug field latch_count redundant.
The field latch_owner must be preserved, because
fil_space_t::is_owner() is being used in all builds.
lock_sys_t::latch: Use srw_lock_debug in debug builds.
This makes the debug fields writer, readers redundant.
lock_sys_t::is_holder(): A new debug predicate to check if
the current thread is holding lock_sys.latch in any mode.
trx_rseg_t::latch: Use srw_lock_debug in debug builds.
When innodb_undo_log_truncate=ON causes an InnoDB undo tablespace
to be truncated, we must guarantee that the undo tablespace will
be rebuilt atomically: After mtr_t::commit_shrink() has durably
written the mini-transaction that rebuilds the undo tablespace,
we must not write any old pages to the tablespace.
To guarantee this, in trx_purge_truncate_history() we used to
traverse the entire buf_pool.flush_list in order to acquire
exclusive latches on all pages for the undo tablespace that
reside in the buffer pool, so that those pages cannot be written
and will be evicted during mtr_t::commit_shrink(). But, this
traversal may interfere with the page writing activity of
buf_flush_page_cleaner(). It would be better to lazily discard
the old pages of the truncated undo tablespace.
fil_space_t::is_being_truncated, fil_space_t::clear_stopping(): Remove.
fil_space_t::create_lsn: A new field, identifying the LSN of the
latest rebuild of a tablespace.
buf_page_t::flush(), buf_flush_try_neighbors(): Evict pages whose
FIL_PAGE_LSN is below fil_space_t::create_lsn.
mtr_t::commit_shrink(): Update fil_space_t::create_lsn and
fil_space_t::size right before the log is durably written and the
tablespace file is being truncated.
fsp_page_create(), trx_purge_truncate_history(): Simplify the logic.
Reviewed by: Thirunarayanan Balathandayuthapani, Vladislav Lesin
Performance tested by: Axel Schwenke
Correctness tested by: Matthias Leich
mtr_t::commit(): If IMPORT TABLESPACE is first-time-dirtying blocks,
acquire both log_sys.mutex and log_sys.flush_order_mutex to assign
a valid m_commit_lsn so that the block will be inserted into the
correct position of buf_pool.flush_list.
This fixes occasional debug assertion failures when running the
regression test suite.
Reviewed by: Vladislav Lesin
The log_sys.lsn_lock that was introduced in
commit a635c40648
had better be located in the same cache line with log_sys.latch
so that log_t::append_prepare() needs to modify only two first
cache lines where log_sys is stored.
log_t::lsn_lock: On Linux, change the type from pthread_mutex_t to
something that may be as small as 32 bits, to pack more data members
in the same cache line. On Microsoft Windows, CRITICAL_SECTION works
better.
log_t::check_flush_or_checkpoint_: Renamed to need_checkpoint.
There is no need to pause all writer threads in log_free_check() when
we only need to write log_sys.buf to ib_logfile0. That will be done in
mtr_t::commit().
log_t::append_prepare_wait(): Make the member function non-static
to simplify the call interface, and add a parameter for the LSN.
log_t::append_prepare(): Invoke append_prepare_wait() at most once.
Only set_check_for_checkpoint() if a log checkpoint needs to
be written. If the log buffer needs to be written, we will take care
of it ourselves later in our caller. This will reduce interference
with log_free_check() in other threads.
mtr_t::commit(): Call log_write_up_to() if needed.
log_t::get_write_target(): Return a log_write_up_to() target
to mtr_t::commit().
buf_flush_ahead(): If we are in furious flushing, call
log_sys.set_check_for_checkpoint() so that all writers will wait
in log_free_check() until the checkpoint is done. Otherwise,
the test innodb.insert_into_empty could occasionally report
an error "Crash recovery is broken".
log_check_margins(): Replaced by log_free_check().
log_flush_margin(): Removed. This is part of mtr_t::commit()
and other operations that write log.
log_t::create(), log_t::attach(): Guarantee that buf_free < max_buf_free
will always hold on PMEM, to satisfy an assumption of
log_t::get_write_target().
log_write_up_to(): Assert lsn!=0. Such calls are not incorrect, but it
is cheaper to test that single unlikely condition in mtr_t::commit()
rather than test several conditions in log_write_up_to().
innodb_drop_database(), unlock_and_close_files(): Check the LSN before
calling log_write_up_to().
ha_innobase::commit_inplace_alter_table(): Remove redundant calls to
log_write_up_to() after calling unlock_and_close_files().
Reviewed by: Vladislav Vaintroub
Stress tested by: Matthias Leich
Performance tested by: Steve Shaw
- The 11.2 test innodb.sys_truncate_debug fails while executing insert statement.
Reason for the failure is that same mini-transaction does freeing, allocating
and freeing the same page. Page initialization clears the FIL_PAGE_LSN
on the page, fails to set the FIL_PAGE_LSN after freeing the same page.
This issue is caused by commit f46efb4476
mtr_t::commit(): Should set the FIL_PAGE_LSN even though page is freed
InnoDB was violating the write-ahead-logging protocol when a file
was being deleted, like this:
1. fil_delete_tablespace() set the fil_space_t::STOPPING flag
2. The buf_flush_page_cleaner() thread discards some changed pages for
this tablespace advances the log checkpoint a little.
3. The server process is killed before fil_delete_tablespace() wrote
a FILE_DELETE record.
4. Recovery will try to apply log to pages of the tablespace, because
there was no FILE_DELETE record. This will fail, because some pages
that had been modified since the latest checkpoint had not been written
by the page cleaner.
Page writes must not be stopped before a FILE_DELETE record has been
durably written.
fil_space_t::drop(): Replaces fil_space_t::check_pending_operations().
Add the parameter detached_handle, and return a tablespace pointer
if this thread was the first one to stop I/O on the tablespace.
mtr_t::commit_file(): Remove the parameter detached_handle, and
move some handling to fil_space_t::drop().
fil_space_t: STOPPING_READS, STOPPING_WRITES: Separate flags for STOPPING.
We want to stop reads (and encryption) before stopping page writes.
fil_space_t::is_stopping_writes(), fil_space_t::get_for_write():
Special accessors for the write path.
fil_space_t::flush_low(): Ignore the STOPPING_READS flag and only
stop if STOPPING_WRITES is set, to avoid an infinite loop in
fil_flush_file_spaces(), which was occasionally repeated by
running the test encryption.create_or_replace.
Reviewed by: Vladislav Lesin
Tested by: Matthias Leich
buf_page_free(): Flag the freed page as modified if it is found in
the buffer pool.
buf_flush_page(): If the page has been freed, ensure that the log
for it has been durably written, before removing the page
from buf_pool.flush_list.
FindBlockX: Find also MTR_MEMO_PAGE_X_MODIFY in order to avoid an
occasional failure of innodb.innodb_defrag_concurrent, which involves
freeing and reallocating pages in the same mini-transaction.
This fixes a regression that was introduced in
commit a35b4ae898 (MDEV-15528).
This logic was tested by commenting out the $shutdown_timeout line
from a test and running the following:
./mtr --rr innodb.scrub
rr replay var/log/mysqld.1.rr/mariadbd-0
A breakpoint in the modified buf_flush_page() was hit, and the
FIL_PAGE_LSN of that page had been last modified during the
mtr_t::commit() of a mini-transaction where buf_page_free()
had been executed on that page.
The MDEV-29693 conflict resolution is from Monty, as well as is
a bug fix where ANALYZE TABLE wrongly built histograms for
single-column PRIMARY KEY.
Also includes a fix for safe_malloc error reporting.
Other things:
- Copied main.log_slow from 10.4 to avoid mtr issue
Disabled test:
- spider/bugfix.mdev_27239 because we started to get
+Error 1429 Unable to connect to foreign data source: localhost
-Error 1158 Got an error reading communication packets
- main.delayed
- Bug#54332 Deadlock with two connections doing LOCK TABLE+INSERT DELAYED
This part is disabled for now as it fails randomly with different
warnings/errors (no corruption).
- InnoDB fails to mark the page status as FREED during freeing
of page for temporary tablespace. This behaviour affects
scrubbing and doesn't write all zeroes in file even though
pages are freed.
mtr_t::free(): Mark the page as freed for temporary tablespace
also
The new statistics is enabled by adding the "engine", "innodb" or "full"
option to --log-slow-verbosity
Example output:
# Pages_accessed: 184 Pages_read: 95 Pages_updated: 0 Old_rows_read: 1
# Pages_read_time: 17.0204 Engine_time: 248.1297
Page_read_time is time doing physical reads inside a storage engine.
(Writes cannot be tracked as these are usually done in the background).
Engine_time is the time spent inside the storage engine for the full
duration of the read/write/update calls. It uses the same code as
'analyze statement' for calculating the time spent.
The engine statistics is done with a generic interface that should be
easy for any engine to use. It can also easily be extended to provide
even more statistics.
Currently only InnoDB has counters for Pages_% and Undo_% status.
Engine_time works for all engines.
Implementation details:
class ha_handler_stats holds all engine stats. This class is included
in handler and THD classes.
While a query is running, all statistics is updated in the handler. In
close_thread_tables() the statistics is added to the THD.
handler::handler_stats is a pointer to where statistics should be
collected. This is set to point to handler::active_handler_stats if
stats are requested. If not, it is set to 0.
handler_stats has also an element, 'active' that is 1 if stats are
requested. This is to allow engines to avoid doing any 'if's while
updating the statistics.
Cloned or partition tables have the pointer set to the base table if
status are requested.
There is a small performance impact when using --log-slow-verbosity=engine:
- All engine calls in 'select' will be timed.
- IO calls for InnoDB reads will be timed.
- Incrementation of counters are done on local variables and accesses
are inline, so these should have very little impact.
- Statistics has to be reset for each statement for the THD and each
used handler. This is only 40 bytes, which should be neglectable.
- For partition tables we have to loop over all partitions to update
the handler_status as part of table_init(). Can be optimized in the
future to only do this is log-slow-verbosity changes. For this to work
we have to update handler_status for all opened partitions and
also for all partitions opened in the future.
Other things:
- Added options 'engine' and 'full' to log-slow-verbosity.
- Some of the new files in the test suite comes from Percona server, which
has similar status information.
- buf_page_optimistic_get(): Do not increment any counter, since we are
only validating a pointer, not performing any buf_pool.page_hash lookup.
- Added THD argument to save_explain_data_intern().
- Switched arguments for save_explain_.*_data() to have
always THD first (generates better code as other functions also have THD
first).
fil_delete_tablespace() stores file handle in local variable and calls
mtr_t::commit_file()=>fil_system_t::detach(..., detach_handle=true), which
sets space->chain.start->handle = OS_FILE_CLOSED. fil_system_t::detach()
is invoked under fil_system.mutex.
But before the mutex is acquired some parallel thread can change
space->chain.start->handle. fil_delete_tablespace() returns value, stored
in local variable, i.e. wrong value.
File handle can be closed, for example, from buf_flush_space() when the
limit of innodb_open_files exceded and fil_space_t::get() causes
fil_space_t::try_to_close() call.
fil_space_t::try_to_close() is executed under fil_system.mutex. And
mtr_t::commit_file() locks it for fil_system_t::detach() call.
fil_system_t::detach() returns detached file handle if its argument
detach_handle is true. The fix is to let mtr_t::commit_file() to pass
that detached file handle to fil_delete_tablespace().
This is a partial revert of
commit 8b6a308e46 (MDEV-29883)
and a follow-up to the
merge commit 394fc71f4f (MDEV-24569).
The latching order related to any operation that accesses the allocation
metadata of an InnoDB index tree is as follows:
1. Acquire dict_index_t::lock in non-shared mode.
2. Acquire the index root page latch in non-shared mode.
3. Possibly acquire further index page latches. Unless an exclusive
dict_index_t::lock is held, this must follow the root-to-leaf,
left-to-right order.
4. Acquire a *non-shared* fil_space_t::latch.
5. Acquire latches on the allocation metadata pages.
6. Possibly allocate and write some pages, or free some pages.
btr_get_size_and_reserved(), dict_stats_update_transient_for_index(),
dict_stats_analyze_index(): Acquire an exclusive fil_space_t::latch
in order to avoid a deadlock in fseg_n_reserved_pages() in case of
concurrent access to multiple indexes sharing the same "inode page".
fseg_page_is_allocated(): Acquire an exclusive fil_space_t::latch
in order to avoid deadlocks. All callers are holding latches
on a buffer pool page, or an index, or both.
Before commit edbde4a11f (MDEV-24167)
a third mode was available that would not conflict with the shared
fil_space_t::latch acquired by ha_innobase::info_low(),
i_s_sys_tablespaces_fill_table(),
or i_s_tablespaces_encryption_fill_table().
Because those calls should be rather rare, it makes sense to use
the simple rw_lock with only shared and exclusive modes.
fil_crypt_get_page_throttle(): Avoid invoking fseg_page_is_allocated()
on an allocation bitmap page (which can never be freed), to avoid
acquiring a shared latch on top of an exclusive one.
mtr_t::s_lock_space(), MTR_MEMO_SPACE_S_LOCK: Remove.
mtr_t::commit(): Add special handling of
innodb_immediate_scrub_data_uncompressed for TEMPORARY TABLE.
This fixes a regression that was caused by
commit de4030e4d4 (MDEV-30400).
