It may be beneficial for slow-path, where new thread is to be created.
However it is harmful for fast-path, where thread cache can fulfil
request: it caused 2 extra loads of write-contended variables.
Also shrink variables scope.
Part of MDEV-19515 - Improve connect speed
Removed FD_CLOEXEC setting: already done by mysql_socket_accept().
Moved O_NONBLOCK setting out of accepting loop.
Removed blocking acceptance attempt, which was dead code. It was supposed
to be executed before the last iteration, however it was actually executed
during the last iteration. And it is not correct to block on one socket
leaving other sockets unattended anyway.
Gives ~5% throughput improvemet in sysbench connect benchmark.
Part of MDEV-19515 - Improve connect speed
abort_loop is intended to break accepting connections loop in main thread,
however it is being used for other purposes.
cache_thread() is governed by kill_cached_threads, no need to check
abort_loop here.
Check in create_new_thread() is redundant, abort_loop already checked
by caller.
Part of MDEV-19515 - Improve connect speed
This commit adds a new feature to the server to add comments at the database
level. 1024 bytes is the maximum comment length allowed. If the comment length
exceeds this limit, a new error/warning code 4144 is thrown, based on whether
thd->is_strict_mode() is true/false. The database comment is also added to the
db.opt file, as well as to the information_schema.schemata table.
In wsrep_plugins_post_init we iterate all theads and if they are
galera appliers (wsrep_applier) we init session variables. However,
current_thd was not set and recent changes on session variables
require holding LOCK_gloal_system_variables mutex.
This is 10.4 version.
With MDEV-19384 fixed FTWRL releases HANDLER locks early, which allows
concurrent threads to go. Test case may get stuck on FTWRL waiting for
LOCK TABLES.
So to push index condition for each join tab we have calculate the index condition that can be pushed and then
remove this index condition from the original condition. This is done through the function make_cond_remainder.
The problem is the function make_cond_remainder does not remove index condition when there is an OR operator.
Fixed this by making the function make_cond_remainder to keep in mind of the OR operator.
Also updated results for multiple test files which were incorrectly updated by the commit e0c1b3f242
code which was supposed to remove the condition present in the index
condition was not getting executed when the condition had OR operator, with AND the pushed
index condition was getting removed from where.
This problem affects all versions starting from 5.5 but this is a performance improvement, so fixing it in 10.4
For a table column `a`, the above expressions logically
equate to false in all cases.
With this patch the optimizer knows about this and queries
like:
SELECT * FROM t1 WHERE a!=a
no longer need to evaluate a!=a for every row.
The same applies if the expression was `a<a`, or `a>a`
An `EXPLAIN SELECT COOUNT(*) FROM t1 WHERE a<a` will show:
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE NULL NULL NULL NULL NULL NULL NULL Impossible WHERE
Similarly `NOT (a!=a)` is always true.
EXPLAIN SELECT COUNT(*) FROM t1 WHERE not (a!=a);
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE NULL NULL NULL NULL NULL NULL NULL Select tables optimized away
With MAX_INDEXIES=64(default), key_map=Bitmap<64> is just a wrapper around
ulonglong and thus "trivial" (can be bzero-ed, or memcpy-ed, and stays
valid still)
With MAX_INDEXES=128, key_map = Bitmap<128> is not a "trivial" type
anymore. The implementation uses MY_BITMAP, and MY_BITMAP contains pointers
which make Bitmap invalid, when it is memcpy-ed/bzero-ed.
The problem in 10.4 is that there are many new key_map members, inside TABLE
or KEY, and those are often memcopied and bzeroed
The fix makes Bitmap "trivial", by inlining most of MY_BITMAP functionality.
pointers/heap allocations are not used anymore.
Parenthesis around table names and derived tables should be allowed
in FROM clauses and some other context as it was in earlier versions.
Returned test queries that used such parenthesis in 10.3 to their
original form. Adjusted test results accordingly.
The deadlock happened between FTWRL under open HANDLER, LOCK TABLE and
DROP DATABASE
Fixed by reverting the previous fix for handler open in
lock_global_read_lock()
Fixed the original (wrong) test case in flush_read_lock.test to be
repeatable.
Windows does atomic writes, as long as they are aligned and multiple
of sector size. this is documented in MSDN.
Fix innodb.doublewrite test to always use doublewrite buffer,
(even if atomic writes are autodetected)
Regression after reverting fair THD members constructors/destructors.
vars_list can be used standalone, in such cases destructor is needed.
Part of MDEV-14984 - regression in connect performance
The accessor dtuple_get_nth_v_field() was defined differently between
debug and release builds in MySQL 5.7.8 in
mysql/mysql-server@c47e1751b7
and a debug assertion to document or enforce the questionable assumption
tuple->v_fields == &tuple->fields[tuple->n_fields] was missing.
This was apparently no problem until MDEV-11369 introduced instant
ADD COLUMN to MariaDB Server 10.3. With that work present, in one
test case, trx_undo_report_insert_virtual() could in release builds
fetch the wrong value for a virtual column.
We replace many of the dtuple_t accessors with const-preserving
inline functions, and fix missing or misleadingly applied const
qualifiers accordingly.
log_checkpoint(), log_make_checkpoint_at(): Remove the parameter
write_always. It seems that the primary purpose of this parameter
was to ensure in the function recv_reset_logs() that both checkpoint
header pages will be overwritten, when the function is called from
the never-enabled function recv_recovery_from_archive_start().
create_log_files(): Merge recv_reset_logs() to its only caller.
Debug instrumentation: Prefer to flush the redo log, instead of
triggering a redo log checkpoint.
page_header_set_field(): Disable a debug assertion that will
always fail due to MDEV-19344, now that we no longer initiate
a redo log checkpoint before an injected crash.
In recv_reset_logs() there used to be two calls to
log_make_checkpoint_at(). The apparent purpose of this was
to ensure that both InnoDB redo log checkpoint header pages
will be initialized or overwritten.
The second call was removed (without any explanation) in MySQL 5.6.3:
mysql/mysql-server@4ca37968da
In MySQL 5.6.8 WL#6494, starting with
mysql/mysql-server@00a0ba8ad9
the function recv_reset_logs() was not only invoked during
InnoDB data file initialization, but also during a regular
startup when the redo log is being resized.
mysql/mysql-server@45e9167983
in MySQL 5.7.2 removed the UNIV_LOG_ARCHIVE code, but still
did not remove the parameter write_always.
Originally introduced by e972125f1 to avoid harmless wait for
LOCK_global_system_variables in a newly created thread, which creation was
initiated by system variable update.
At the same time it opens dangerous hole, when system variable update
thread already released LOCK_global_system_variables and ack_receiver
thread haven't yet completed new THD construction. In this case THD
constructor goes completely unprotected.
Since ack_receiver.stop() waits for the thread to go down, we have to
temporarily release LOCK_global_system_variables so that it doesn't
deadlock with ack_receiver.run(). Unfortunately it breaks atomicity
of rpl_semi_sync_master_enabled updates and makes them not serialized.
LOCK_rpl_semi_sync_master_enabled was introduced to workaround the above.
TODO: move ack_receiver start/stop into repl_semisync_master
enable_master/disable_master under LOCK_binlog protection?
Part of MDEV-14984 - regression in connect performance
Only take LOCK_plugin for plugin system variables.
Reverted optimisation that was originally done for session tracker: it
makes much less sense now. Specifically only if connections would want to
track plugin session variables changes and these changes would actually
happen frequently. If this ever becomes an issue, there're much better
ways to optimise this workload.
Part of MDEV-14984 - regression in connect performance
Rather than parsing session_track_system_variables when thread starts, do
it when first trackable event occurs.
Benchmarked on a 2socket/20core/40threads Broadwell system using sysbench
connect brencmark @40 threads (with select 1 disabled):
101379.77 -> 143016.68 CPS, whereas 10.2 is currently at 137766.31 CPS.
Part of MDEV-14984 - regression in connect performance
When enabling system variables tracker, make a copy of
global_system_variables.session_track_system_variables
under LOCK_global_system_variables. This protects from concurrent variable
updates and potential freed memory access, as well as from variable
reconstruction (which was previously protected by LOCK_plugin).
We can also use this copy as a session variable pointer, so that we don't
have to allocate memory and reconstruct it every time it is referenced.
For this very reason we don't need buffer_length stuff anymore.
As well as don't we need to take LOCK_plugin early in ::enable().
Unified ::parse_var_list() to acquire LOCK_plugin unconditionally.
For no apparent reason it wasn't previously acquired for global
variable update.
Part of MDEV-14984 - regression in connect performance
One less new/delete per connection.
Removed m_mem_flag since most allocs are thread specific. The only
exception are allocs performed during initialization.
Removed State_tracker and Session_tracker constructors as they don't make
sense anymore.
No reason to access session_sysvars_tracker via get_tracker(), so access
it directly instead.
Part of MDEV-14984 - regression in connect performance
tool_list is a temporary list needed only for SET SESSION
session_track_system_variables, so allocate it on stack instead.
Sane reinit() calls.
Saves 2 new/delete per connection.
Part of MDEV-14984 - regression in connect performance
