* it isn't "pfs" function, don't call it Item_func_pfs,
don't use item_pfsfunc.*
* tests don't depend on performance schema, put in the main suite
* inherit from Item_str_ascii_func
* use connection collation, not utf8mb3_general_ci
* set result length in fix_length_and_dec
* do not set maybe_null
* use my_snprintf() where possible
* don't set m_value.ptr on every invocation
* update sys schema to use the format_pico_time()
* len must be size_t (compilation error on Windows)
* the correct function name for double->double is fabs()
* drop volatile hack
AWK in used in Debian SysV-init and postinst scripts to determine
is there enough space starting MariaDB database or create new
database to target destination.
These AWK scripts can be rewrited to use pure SH or help
using Coreutils which is mandatory for usage of MariaDB currently.
Reasoning behind this is to get rid of one very less used dependency
Commit reduces need of AWK-command at least
for Debian mariadb-server-compat package.
Commit removes need of AWK-command from
scripts/mysql_install_db.sh script.
AWK command is replace by purely Posix sh compiliant version.
In commit d6aed21621 a condition at
the start of buf_read_ahead_random() was refactored. Only the caller
buf_read_recv_pages() was adjusted for this. We must in fact adjust
every caller and make sure that spare blocks will be allocated
while crash recovery is in progress. This is the simplest fix;
ideally recovery would operate on the compressed page frame.
The observed recovery hang occurred because pages 0 and 3 of a
tablespace were being read due to buf_page_get_gen() calls by
trx_resurrect_table_locks() before the log records for these pages
had been applied. In buf_page_t::read_complete() we would skip
the call to recv_recover_page() because no uncompressed page frame
had been allocated for the block.
In some cases, the errors would not be written to the log.
This was however not critical as in most cases mysql_install_db
should not normally write anything to the log.
- InnoDB does rollback the whole transaction and discards the
savepoint when there is a failure happens during bulk
insert operation. When server request to release the savepoint,
InnoDB should return DB_SUCCESS when it deals with bulk
insert operation
The failed assertion was about encountering the same rowid value in
two different partitions.
This wasn't possible with InnoDB previously: InnoDB used a global counter
to produce rowid values for hidden PK.
After the fix for MDEV-19506, it uses per-table counters so it's easily
possible to get the same hidden PK values in different tables.
The crash happened due to rows=2 vs rows=1 difference between how the
estimate of number of rows in a derived table is computed in
TABLE_LIST::fetch_number_of_rows() and JOIN::add_keyuses_for_splitting().
Made JOIN::add_keyuses_for_splitting() use the result of computations in
TABLE_LIST::fetch_number_of_rows().
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
lock_sec_rec_some_has_impl(): Remove a harmful condition that caused the
performance regression and should not have been added in
commit b6e41e3872 in the first place.
Locking transactions that have not modified any persistent tables
can carry the transaction identifier 0.
trx_t::max_inactive_id: A cache for trx_sys_t::find_same_or_older().
The value is not reset on transaction commit so that previous results
can be reused for subsequent transactions. The smallest active
transaction ID can only increase over time, not decrease.
trx_sys_t::find_same_or_older(): Remember the maximum previous id for which
rw_trx_hash.iterate() returned false, to avoid redundant iterations.
lock_sec_rec_read_check_and_lock(): Add an early return in case we are
already holding a covering table lock.
lock_rec_convert_impl_to_expl(): Add a template parameter to avoid
a redundant run-time check on whether the index is secondary.
lock_rec_convert_impl_to_expl_for_trx(): Move some code from
lock_rec_convert_impl_to_expl(), to reduce code duplication due
to the added template parameter.
Reviewed by: Vladislav Lesin
Tested by: Matthias Leich
This is a follow-up to
commit de4030e4d4 (MDEV-30400),
which fixed some hangs related to B-tree split or merge.
btr_root_block_get(): Use and update the root page guess. This is just
a minor performance optimization, not affecting correctness.
btr_validate_level(): Remove the parameter "lockout", and always
acquire an exclusive dict_index_t::lock in CHECK TABLE without QUICK.
This is needed in order to avoid latching order violation in
btr_page_get_father_node_ptr_for_validate().
btr_cur_need_opposite_intention(): Return true in case
btr_cur_compress_recommendation() would hold later during the
mini-transaction, or if a page underflow or overflow is possible.
If we return true, our caller will escalate to aqcuiring an exclusive
dict_index_t::lock, to prevent a latching order violation and deadlock
during btr_compress() or btr_page_split_and_insert().
btr_cur_t::search_leaf(), btr_cur_t::open_leaf():
Also invoke btr_cur_need_opposite_intention() on the leaf page.
btr_cur_t::open_leaf(): When escalating to exclusive index locking,
acquire exclusive latches on all pages as well.
innobase_instant_try(): Return an error code if the root page cannot
be retrieved.
In addition to the normal stress testing with Random Query Generator (RQG)
this has been tested with
./mtr --mysqld=--loose-innodb-limit-optimistic-insert-debug=2
but with the injection in btr_cur_optimistic_insert() for non-leaf pages
adjusted so that it would use the value 3. (Otherwise, infinite page
splits could occur in some mtr tests.)
Tested by: Matthias Leich
srv_start(): If we are going to close the log file in
mariadb-backup --prepare, call buf_flush_sync() before
calling recv_sys.debug_free() to ensure that the log file
will not be accessed.
This fixes a rather rare failure in the test
mariabackup.innodb_force_recovery where buf_flush_page_cleaner()
would invoke log_checkpoint_low() because !recv_recovery_is_on()
would hold due to the fact that recv_sys.debug_free() had
already been called. Then, the log write for the checkpoint
would fail because srv_start() had invoked log_sys.log.close_file().
Building of MariaDB server version 11.0 and 11.1 fails
on MacOS 12.x (Monterey). Build failure happened on generating
header files from the error messages contained in the file
errmsg-utf8.txt. This process is performed by the utility comp_err
that crashes when it is run on MacOS Monterey.
comp_err invokes my_init at the very beginning of start before
initialization of thread environment done. While executing my_init
the function my_readlink is called. my_readlink is wrapper around
the system call readlink with extra errors handling. In case the
system call readlink returns error the following block of code is run
if (my_thread_var)
my_errno= errno;
my_thred_var is macros that expanded to invocation of
my_pthread_getspecific() against supplied thread specific key
THR_KEY_mysys. Unfortunately, the tsd key THR_KEY_mysys is initialized
right after the call of my_init() so return value of pthread_getspecific
is platform dependent. On Linux pthread_getspecific returns NULL
if key is not a valid TSD key. On MacOS, the effect of calling
pthread_getspecific() with a key value not obtained from pthread_key_create()
is undefined. So, on MacOS pthread_getspecific() returns some invalid address
where the errno value is written. It leads to a crash latter when the library
API function pthread_self() is called.
To fix the issue, initialization of thread environment is moved at very
beginning of the main() function in order to run it as the first step
to have full initiliazed environment at the moment when my_init()
is ivoked.
Do not set any flags in the items for constant subformulas TRUE/FALSE when
checking pushability of a formula into a view. Occurrences of these
subformulas can be ignored when checking pushability of the formula.
At the same time the items used for these constants became immutable
starting from version 10.7.
Approved by Oleksandr Byelkin <sanja@mariadb.com>
Type_handler::partition_field_append_value() erroneously
passed the address of my_collation_contextually_typed_binary
to conversion functions copy_and_convert() and my_convert().
This happened because generate_partition_syntax_for_frm()
was called from mysql_create_frm_image() in the stage when
the fields in List<Create_field> can still contain unresolved
contextual collations, like "binary" in the reported crash scenario:
ALTER TABLE t CHANGE COLUMN a a CHAR BINARY;
Fix:
1. Splitting mysql_prepare_create_table() into two parts:
- mysql_prepare_create_table_stage1() interates through
List<Create_field> and calls Create_field::prepare_stage1(),
which performs basic attribute initialization, including
context collation resolution.
- mysql_prepare_create_table_finalize() - the rest of the
old mysql_prepare_create_table() code.
2. Changing mysql_create_frm_image():
It now calls:
- mysql_prepare_create_table_stage1() in the very
beginning, before the partition related code.
- mysql_prepare_create_table_finalize() in the end,
instead of the old mysql_prepare_create_table() call
3. Adding mysql_prepare_create_table() as a wrapper
for two calls:
mysql_prepare_create_table_stage1() ||
mysql_prepare_create_table_finalize()
so the code stays unchanged in the other places
where mysql_prepare_create_table() was used.
4. Changing prototype for Type_handler::Column_definition_prepare_stage1()
Removing arguments:
- handler *file
- ulonglong table_flags
Adding a new argument instead:
- column_definition_type_t type
This allows to call Column_definition_prepare_stage1() and
therefore to call mysql_prepare_create_table_stage1()
before instantiation of a handler.
This simplifies the code, because in case of a partitioned table,
mysql_create_frm_image() creates a handler of the underlying
partition first, the frees it and created a ha_partition
instance instead.
mysql_prepare_create_table() before the fix was called with the final
(ha_partition) handler.
5. Moving parts of Column_definition_prepare_stage1() which
need a pointer to handler and table_flags to
Column_definition_prepare_stage2().
fil_node_open_file_low() tries to close files from the top of
fil_system.space_list if the number of opened files is exceeded.
It invokes fil_space_t::try_to_close(), which iterates the list searching
for the first opened space. Then it just closes the space, leaving it in
the same position in fil_system.space_list.
On heavy files opening, like during 'SHOW TABLE STATUS ...' execution,
if the number of opened files limit is reached,
fil_space_t::try_to_close() iterates more and more closed spaces before
reaching any opened space for each fil_node_open_file_low() call. What
causes performance regression if the number of spaces is big enough.
The fix is to keep opened spaces at the top of fil_system.space_list,
and move closed files at the end of the list.
For this purpose fil_space_t::space_list_last_opened pointer is
introduced. It points to the last inserted opened space in
fil_space_t::space_list. When space is opened, it's inserted to the
position just after the pointer points to in fil_space_t::space_list to
preserve the logic, inroduced in MDEV-23855. Any closed space is added
to the end of fil_space_t::space_list.
As opened spaces are located at the top of fil_space_t::space_list,
fil_space_t::try_to_close() finds opened space faster.
There can be the case when opened and closed spaces are mixed in
fil_space_t::space_list if fil_system.freeze_space_list was set during
fil_node_open_file_low() execution. But this should not cause any error,
as fil_space_t::try_to_close() still iterates spaces in the list.
There is no need in any test case for the fix, as it does not change any
functionality, but just fixes performance regression.
- Change to use 'mariadbd' instead of 'mysqld' in help texts and other
visible places.
- Start binary 'mariadbd' instead of 'mysqld'. This will remove a warning
in 11.0 when running mysql_install_db.
- Use my_print_defaults --mariadbd instead of --mysqld
- Use --skip-log-error if the user don't have access to log-error file.
This it needed to allow mysql_install_db to work silenty for users that
has not write access to /var/log.
Other things:
- Updated my_print_defaults to support --mariadbd
Charset names in the 'languages' line are not used any more.
Removing to avoid confusion.
All messages in errmsg-utf8.txt are in utf8 now.
Charset names should have been removed in MySQL-5.5 during: https://dev.mysql.com/worklog/task/?id=751
Bump version number.
While downgrades are not supported and misguided attempts at it could
cause serious corruption especially after
commit b07920b634
it might be useful if InnoDB would start up even after an upgrade to
MariaDB Server 11.0 or later had removed the change buffer.
innodb_change_buffering_update(): Disallow anything else than
innodb_change_buffering=none when the change buffer is corrupted.
ibuf_init_at_db_start(): Mention a possible downgrade in the corruption
error message. If innodb_change_buffering=none, ignore the error but do
not initialize ibuf.index.
ibuf_free_excess_pages(), ibuf_contract(), ibuf_merge_space(),
ibuf_update_max_tablespace_id(), ibuf_delete_for_discarded_space(),
ibuf_print(): Check for !ibuf.index.
ibuf_check_bitmap_on_import(): Remove some unnecessary code.
This function is only accessing change buffer bitmap pages in a
data file that is not attached to the rest of the database.
It is not accessing the change buffer tree itself, hence it does
not need any additional mutex protection.
This has been tested both by starting up MariaDB Server 10.8 on
a 11.0 data directory, and by running ./mtr --big-test while
ibuf_init_at_db_start() was tweaked to always fail.
The mini-benchmark.sh script failed to run in the latest Fedora
distributions in GitLab CI. It requires `lscpu` resolved by installing
util-linux.
Additionally, executing the benchmark inside a Docker container had
failed because of increased Docker security in recent updates. In
particular the `renice` and `taskset` operations are not permitted.
Neither are the required `perf` operations.
https://docs.docker.com/engine/security/seccomp/
Allow these operations to fail gracefully, and test then skip `perf`,
allowing the remaining benchmark activities to proceed.
Other minor changes to the CI are included such as allowing sanitizer
jobs to fail and using "needs" in the mini-benchmark pipeline.
All new code of the whole pull request, including one or several files
that are either new files or modified ones, are contributed under the
BSD-new license. I am contributing on behalf of my employer
Amazon Web Services, Inc.
Removed an old '* 2' from the HASH join cost. This was made obsolete by
a later patch that added cost for copying the data out from the join buffer
to table->record.
I also added some 'echo' to some test cases to make it easier to debug
test case changes.
Test case changes:
- subselect3_jcl6 and subselect_sj2_jcl6 result changes as materialized
tables changed to hash join + first_match
Firstmatch_picker::check_qep() has an optimization that allows firstmatch
to be used together with join buffer under some conditions. In this
case the cost was assumed to be same as what best_access_path()
had calculated.
However if HASH+join_buffer was used, then
fix_semijoin_strategies_for_picked_join_order() would remove the
join_buffer (which would cause a full join to be used) and the cost
assumption by Firstmatch_picker::check_qep() would be wrong.
Later check_join_cache_usage() sees that it's a full scan and decides
it can use join buffering, (But not the hash join).
Fixed by also allowing HASH joins with firstmatch.
This removes the need to change disable and re-enable join buffer.
Test case changes:
- HASH join used with firstmatch (Using join buffer (flat, BNLH join))
- Filtered could change with firstmatch as the conversion with and without
join_buffered lost the filtering information.
- The not "re-enabling join buffer" is shown in main.optimizer_trace
Original code by Sergei, optimized by Monty.
Author: Sergei Petrunia <sergey@mariadb.com>, monty@mariadb.org
