The --gtid-ignore-duplicates option was not working correctly with row-based
replication. When a row event was completed, but before committing, there
was a small window where another multi-source SQL thread could wrongly try
to re-execute the same transaction, without properly ignoring the duplicate
GTID. This would lead to duplicate key error or out-of-order GTID error or
similar.
Thanks to Matt Neth for reporting this and giving an easy way to reproduce
the issue.
Split first_breadth_first_tab() into
JOIN::first_breadth_first_optimization_tab() and
JOIN::first_breadth_first_execution_tab().
This allows to eliminate function call and one condition. Adjusted callers
accordingly.
Overhead change:
first_breadth_first_tab() 0.07% -> out of radar
next_breadth_first_tab() 0.04% -> 0.04%
JOIN::cleanup() 0.15% -> 0.11%
JOIN::save_explain_data_intern() 0.28% -> 0.24%
Moved Apc_target::destroy(), Apc_target::enable() and Apc_targe::disable()
definitions to my_apc.h so that they can be inlined.
Apc_targe::disable() now calls Apc_target::process_apc_requests() only if
there're APC requests. This saves one pthread_mutex_lock() call.
Overhead change:
Apc_target::disable 0.04% -> out of radar
Apc_target::enable 0.03% -> out of radar
Apc_target::process_apc_requests 0.02% -> out of radar
pthread_mutex_lock 0.43% -> 0.42%
pthread_mutex_unlock 0.26% -> 0.25%
Handle the case where the optimizer decides to use
handler->delete_all_rows(), but then this call returns
HA_ERR_UNSUPPORTED and execution switches to regular
row-by-row deletion.
Do not call handler::rebind_psi() and handler::unbind_psi() when performance
schema is compiled out.
Overhead change:
handler::rebind_psi 0.04% -> out of radar
handler::unbind_psi 0.03% -> out of radar
open_table 0.21% -> 0.18%
close_thread_table 0.05% -> 0.05%
Initialize abs_timeout when it is about to be used. This saves one my_hrtime()
call on hot path (when we acquire MDL lock without waiting).
When filling I_S.PROCESSLIST use THD::start_utime/THD::utime_after_query instead
of THD::start_time. This allows us to save 2 clock_gettime() calls.
Overhead change:
__clock_gettime 0.13% -> 0.11% (122 -> 76 calls per OLTP RO transaction)
my_interval_timer 0.07% -> 0.06%
my_hrtime 0.04% -> 0.01%
Pass THD to find_all_keys() and Item_equal::Item_equal().
In MRR use table->in_use instead of current_thd.
This reduces number of pthread_getspecific() calls from 354 to 320.
in ha_delete_table()
* only convert ENOENT and HA_ERR_NO_SUCH_TABLE to warnings
* only return real error codes (that is, not ENOENT and
not HA_ERR_NO_SUCH_TABLE)
* intercept HA_ERR_ROW_IS_REFERENCED to generate backward
compatible ER_ROW_IS_REFERENCED
in mysql_rm_table_no_locks()
* no special code to handle HA_ERR_ROW_IS_REFERENCED
* no special code to handle ENOENT and HA_ERR_NO_SUCH_TABLE
* return multi-table error ER_BAD_TABLE_ERROR <table list> only
when there were many errors, not when there were many
tables to drop (but only one table generated an error)
When RENAME TABLE is executed, it apparently does not check whether the engine
is available (unlike ALTER TABLE .. RENAME, which does). It means that if the
engine in question was not loaded on some reason, the table might become
unusable, since the engine won't know about the change.
With this patch RENAME TABLE fails if storage engine is not available.
when --bind-address is not specificed explicitly (or set to '*')
MariaDB tries all wildcard addresses. Print a warning (not an error)
if a socket cannot be created for some of them.
Still print an error if a socket cannot be created for an address
that a user has specified expicitly with --bind-address.
* take into account that example may be NULL
* use example->safe_charset_converter(), copy-paste from
Item::safe_charset_converter() (example might have its own
implementation)
* handle the case when the charset doesn't need conversion
(and return this).
semisync plugin and setting rpl_semi_sync_master_enabled
There was race condition between INSTALL PLUGIN and SET. It was caused by a
gap in INSTALL PLUGIN when plugin variables were registered but not fully
initialized. Accessing such variables concurrently may reference uninitialized
memory, specifically sys_var_pluginvar::plugin.
Fixed by initializing sys_var_pluginvar::plugin early, before variable is
registered.
semisync plugin and setting rpl_semi_sync_master_enabled
Cleanup:
Removed my_intern_plugin_lock() and my_intern_plugin_lock_ci() wrappers. They
were obsoleted by revision f56dd32bf.
Add log_bin_index, log_bin_basename and relay_log_basename system
variables. Also, convert relay_log_index system variable to
NO_CMD_LINE and implement --relay-log-index as a command line
option.
"Range Checked for Each Record" should be only employed when the other
option would be cross-product join (i.e. the other option is so bad that
we hardly risk anything).
Previous logic was: use RCfER if there are no possible quick selects, or
quick select would read > 100 rows. Also, it didn't always work as
expected due to range optimizer changing table->quick_keys and us
looking at sel->quick_keys.
Another angle is that recent versions have enabled use of Join Buffering
in e.g. outer joins. This further reduces the range of cases where RCfER
should be used.
We are still unable to estimate the cost of RCfER with any precision, so
now changing the condition of "no quick select or quick->records> 100"
to a hopefully better condition "no quick select or quick would cost more
than full table scan".
Changing the error message to:
"...from type 'decimal(0,?)/*old*/' to type ' 'decimal(10,7)'..."
So it's now clear that the master data type is OLD decimal.
Removing Item_cache::used_table_map, Item_cache::used_tables() and
Item_cache::set_used_tables(). Using the same inherited from
Item_basic_constant implementations instead.
Fhe GEOMETRY field should be handled just as the BLOB field. So that was fiexed in field_conv.
One additional bug was found and fixed meanwhile - thet the geometry field subtypes
should also be merged for UNION command.