Relax memory barrier for lock_word.
rw_lock_lock_word_decr() - used to acquire rw-lock, thus we only need to issue
ACQUIRE when we succeed locking.
rw_lock_x_lock_func_nowait() - same as above, but used to attempt to acquire
X-lock.
rw_lock_s_unlock_func() - used to release S-lock, RELEASE is what we need here.
rw_lock_x_unlock_func() - used to release X-lock. Ideally we'd need only RELEASE
here, but due to mess with waiters (they must be loaded after lock_word is
stored) we have to issue both ACQUIRE and RELEASE.
rw_lock_sx_unlock_func() - same as above, but used to release SX-lock.
rw_lock_s_lock_spin(), rw_lock_x_lock_func(), rw_lock_sx_lock_func() -
fetch-and-store to waiters has to issue only ACQUIRE memory barrier, so that
waiters are stored before lock_word is loaded.
Note that there is violation of RELEASE-ACQUIRE protocol here, because we do
on lock:
my_atomic_fas32_explicit((int32*) &lock->waiters, 1, MY_MEMORY_ORDER_ACQUIRE);
my_atomic_load32_explicit(&lock->lock_word, MY_MEMORY_ORDER_RELAXED);
on unlock
my_atomic_add32_explicit(&lock->lock_word, X_LOCK_DECR, MY_MEMORY_ORDER_ACQ_REL);
my_atomic_load32_explicit((int32*) &lock->waiters, MY_MEMORY_ORDER_RELAXED);
That is we kind of synchronize ACQUIRE on lock_word with ACQUIRE on waiters.
It was there before this patch. Simple fix may have negative performance impact.
Proper fix requires refactoring of lock_word.
Relax memory barrier for waiters: these 2 stores must be completed before
os_event_set() finishes. This is guaranteed by RELEASE barrier issued by
mutex.exit() of os_event_set().
Remove volatile modifier from waiters: it's not supposed for inter-thread
communication, use appropriate atomic operations instead.
Changed waiters to int32_t, my_atomic friendly type.
- Remove not used thd_rpl_is_parallel()
- Remove not used mysql_notify_thread_having_shared_lock()
- Remove not needed LOCK_thread_count from MYSQL_BIN_LOG::reset_logs()
- LOCK_thread_count is not protecting against rollback, so this
code and comment is not needed
- Remove mutex_locks in slave.cc that are not needed.
Added THD::assert_not_linked() to ensure that it was safe to remove
- Fixed not repeatable test load_data_stmt_view
- Updated binlog_killed to test removal of mutex
(thanks to Andrei Elkin for test)
- More code comments
When logging ROW_T_INSERT or ROW_T_UPDATE records, we did not normalize
the DB_TRX_ID of the current transaction into 0 if the current transaction
had started (modifying other tables) before the ALTER TABLE started.
MDEV-13654 introduced this normalization for ROW_T_DELETE
and for all operations with ADD PRIMARY KEY, in row_log_table_get_pk().
Introduce the debug flag trx_t::persistent_stats to suppress the
assertion for the updates of persistent statistics during fast
shutdown.
dict_stats_exec_sql(): Do execute the statement even though shutdown
has been initiated.
dict_stats_exec_sql(): Expect the caller to always provide a transaction.
Remove some redundant assertions. The caller must hold dict_sys->mutex,
but holding dict_operation_lock is only necessary for accessing
data dictionary tables, which we are not accessing.
dict_stats_save_index_stat(): Acquire dict_sys->mutex
for invoking dict_stats_exec_sql().
dict_stats_save(), dict_stats_update_for_index(), dict_stats_update(),
dict_stats_drop_index(), dict_stats_delete_from_table_stats(),
dict_stats_delete_from_index_stats(), dict_stats_drop_table(),
dict_stats_rename_in_table_stats(), dict_stats_rename_in_index_stats(),
dict_stats_rename_table(): Use a single caller-provided
transaction that is started and committed or rolled back by the caller.
dict_stats_process_entry_from_recalc_pool(): Let the caller provide
a transaction object.
ha_innobase::open(): Pass a transaction to dict_stats_init().
ha_innobase::create(), ha_innobase::discard_or_import_tablespace():
Pass a transaction to dict_stats_update().
ha_innobase::rename_table(): Pass a transaction to
dict_stats_rename_table(). We do not use the same transaction
as the one that updated the data dictionary tables, because
we already released the dict_operation_lock. (FIXME: there is
a race condition; a lock wait on SYS_* tables could occur
in another DDL transaction until the data dictionary transaction
is committed.)
ha_innobase::info_low(): Pass a transaction to dict_stats_update()
when calculating persistent statistics.
alter_stats_norebuild(), alter_stats_rebuild(): Update the
persistent statistics as well. In this way, a single transaction
will be used for updating the statistics of a whole table, even
for partitioned tables.
ha_innobase::commit_inplace_alter_table(): Drop statistics for
all partitions when adding or dropping virtual columns, so that
the statistics will be recalculated on the next handler::open().
This is a refactored version of Oracle Bug#22469660 fix.
RecLock::add_to_waitq(), lock_table_enqueue_waiting():
Do not allow a lock wait to occur for updating statistics
in a data dictionary transaction, such as DROP TABLE. Instead,
return the previously unused error code DB_QUE_THR_SUSPENDED.
row_merge_lock_table(), row_mysql_lock_table(): Remove dead code
for handling DB_QUE_THR_SUSPENDED.
row_drop_table_for_mysql(), row_truncate_table_for_mysql():
Drop the statistics as part of the data dictionary transaction.
After TRUNCATE TABLE, the statistics will be recalculated on
subsequent ha_innobase::open(), similar to how the logic after
the above-mentioned Oracle Bug#22469660 fix in
ha_innobase::commit_inplace_alter_table() works.
btr_defragment_thread(): Use a single transaction object for
updating defragmentation statistics.
dict_stats_save_defrag_stats(), dict_stats_save_defrag_stats(),
dict_stats_process_entry_from_defrag_pool(),
dict_defrag_process_entries_from_defrag_pool(),
dict_stats_save_defrag_summary(), dict_stats_save_defrag_stats():
Add a parameter for the transaction.
dict_stats_empty_table(): Make public. This will be called by
row_truncate_table_for_mysql() after dropping persistent statistics,
to clear the memory-based statistics as well.
TABLE_SHARE::init_from_binary_frm_image() calls handler_file->index_flags()
before it has set TABLE_SHARE::primary_key (it is 0 while it should be
MAX_KEY in my example).
This causes MyRocks to report wrong index flags (it thinks it's a PK while
it is not), which causes invalid query plans later on.
Do the only thing that seems feasible: adjust field->part_of key to have
correct value in ha_rocksdb::open.
Silence the error log output that was introduced in MySQL 5.7
(MariaDB 10.2.2) if log_warnings=2 or less.
We should still figure out what these messages really indicate
and how to solve the problems.
pc_sleep_if_needed(): Add a parameter for the current time,
so that there will be fewer successive calls to ut_time_ms()
with no I/O between them.
buf_flush_page_cleaner_coordinator(): Exit the first loop
whenever shutdown has been requested. At the start of the loop,
call ut_time_ms() only once. Do not display the message if
log_warnings=2 or less.
* Note: breaking change; since this commit, a plugin that has
worked so far might get rejected due to plugin maturity
* mariabackup is not affected (allows all plugins)
* VERSION file defines SERVER_MATURITY, which defines the
corresponding numeric value as SERVER_MATURITY_LEVEL in
include/mysql_version.h
* The default value for 'plugin_maturity' is SERVER_MATURITY_LEVEL - 1
* Logs a warning if a plugin has maturity lower than
SERVER_MATURITY_LEVEL
* Tests suppress the plugin maturity warning
* Tests use --plugin-maturity=unknown by default so as not to fail
due to the stricter plugin maturity handling
This commit implements aggregate stored functions. The basic idea behind
the feature is:
* Implement a special instruction FETCH GROUP NEXT ROW that will pause
the execution of the stored function. When the instruction is reached,
execution of the initial query resumes "as if" the function returned.
This gives the server the opportunity to advance to the next row in the
result set.
* Stored aggregates behave like regular aggregate functions. The
implementation of thus resides in the class Item_sum_sp. Because it is
an aggregate function, for each new row in the group, the
Item_sum_sp::add() method will be called. This is when execution resumes
and the function does another iteration to "add" one extra element to
the final result.
* When the end of group is reached, val_xxx() method will be called for
the item. This case is handled by another execute step for the stored
function, only with a special flag to force a call to the return
handler. See Item_sum_sp::execute() for details.
To allow this pause and resume semantic, we must preserve the function
context across executions. This is stored in Item_sp::sp_query_arena only for
aggregate stored functions, but has no impact for regular functions.
We also enforce aggregate functions to include the "FETCH GROUP NEXT ROW"
instruction.
Signed-off-by: Vicențiu Ciorbaru <vicentiu@mariadb.org>
In preparation for implementing custom aggregate functions, refactor
the common code between regular stored functions and aggregate stored
functions. This includes:
* initialising SP result field
* executing a SP
* access checks
In addition, refactor sp_head::execute_function to take two extra
parameters, a function rcontext and a Query_arena. These two paremeters
were initially initialised and destroyed within
sp_head::execute_function, but for aggregate stored functions we will
require control over their lifetime. The owner of these objects now
becomes Item_sp.
Signed-off-by: Vicențiu Ciorbaru <vicentiu@mariadb.org>
Problem was that Binlog_checkpoint can happen at random times.
Fixed by not write binlog_checkpoint for the rpl_log test.
Other things:
- Removed not used variable "$keep_gtid_events"
- Added option for show_binlog_events to skip binlog_checkpoint
Includes Spider patches
- 062_mariadb-10.2.0.direct_join_1and3.diff
- 063_mariadb-10.2.0.direct_join_for_single_partition.diff
- Test cases from Kentoku
Allows Spider to push full joins to the Spider engine trough the
create_group_by interface.
Other things:
- Increased MYSQL_VERSION_ID to check for 10211 (latest 10.2 version)
- Fix for const_table at calling create_group_by().
Original author: Kentoku SHIBA
Add support for direct update and direct delete requests for spider.
A direct update/delete request handles all qualified rows in a single
operation rather than one row at a time.
Contains Spiral patches:
006_mariadb-10.2.0.direct_update_rows.diff MDEV-7704
008_mariadb-10.2.0.partition_direct_update.diff MDEV-7706
010_mariadb-10.2.0.direct_update_rows2.diff MDEV-7708
011_mariadb-10.2.0.aggregate.diff MDEV-7709
027_mariadb-10.2.0.force_bulk_update.diff MDEV-7724
061_mariadb-10.2.0.mariadb-10.1.8.diff MDEV-12870
- The differences compared to the original patches:
- Most of the parameters of the new functions are unnecessary. The
unnecessary parameters have been removed.
- Changed bit positions for new handler flags upon consideration of
handler flags not needed by other Spiral patches and handler flags
merged from MySQL.
- Added info_push() (Was originally part of bulk access patch)
- Didn't include code related to handler socket
- Added HA_CAN_DIRECT_UPDATE_AND_DELETE
Original author: Kentoku SHIBA
First reviewer: Jacob Mathew
Second reviewer: Michael Widenius
- Ensure that var= doesn't have a space before =
- Fixed DBUG_PRINT to use %u for unsigned types
- Use "enter" when printing function arguments
- Fixed typos
- Added some extra DBUG_PRINT
- Removed not needed assignment
This is about adding more options to force slave retries
Two new variables has been added:
slave_transaction_retry_errors
- Tells the slave thread to retry transaction for replication when a
query event returns an error from the provided list. Deadlock and
elapsed lock wait timeout errors are automatically added to this list
slave-transaction-retry-interval
- Interval of the slave SQL thread will retry a transaction
in case it failed with a deadlock or elapsed lock wait
timeout or listed in slave_transaction_retry_errors
Other changes:
- Simplifed code for slave_skip_errors (to be aligned with
slave_transaction_retry_errors)
- Renamed print_slave_skip_errors() to make_slave_skip_errors_printable()
- Remove printing error from init_slave_skip_errors as my_bitmap_init()
will do that if needed.
- Generalize has_temporary_error()
Spiral Patch 057: 057_mariadb-10.2.0.partition_index_end.diff MDEV-12999
Original author: Kentoku SHIBA
First reviewer: Jacob Mathew
Second reviewer: Michael Widenius
- In Spider, calling cmp_ref() can be very expensive. In ha_partition.cc
we don't anymore sort rows according to position for the Spider
engine.
- Removed Spider specific call info(HA_EXTRA_STARTING_ORDERED_INDEX_SCAN)
from handle_ordered_index_scan(). It's caused performance issues and
does not change results for queries with ORDER BY.
- The visible effect of this patch is that for some storage engines,
rows may be returned in a different order if there is no ORDER BY clause.
- Based in Spiral Patch 052:
052_mariadb-10.2.0.add_partition_skip_pk_sort_for_non_clustered_index
MDEV-7748
- The major difference from original patch is that there is no variable to
get the old behaviour.
Other things:
- Optimized ha_partition::cmp_ref() and cmp_part_ids() to make them
simpler and faster.
- Changed arguments to cmp_key_part_id() to be same as
cmp_key_rowid_part_id to simplify code.
Original author: Kentoku SHIBA
First reviewer: Jacob Mathew
Second reviewer: Michael Widenius
Contains Spiral patches:
022_mariadb-10.2.0.auto_increment.diff MDEV-7720
030: 030_mariadb-10.2.0.partition_auto_inc_init.diff MDEV-7726
These patches have the following differences compared to the original
patches:
- Added the new #defines for the feature in spd_environ.h instead of in
handler.h because these #defines are needed by Spider and are not needed
by the server.
- Cleaned up code related to the removed variable m_need_info_for_auto_inc
. Changed variable assignment in lock_auto_increment() and
unlock_auto_increment() so that the assignments are done under locks.
- Added a test case.
- Added test result changes resulting from a bug that was fixed by these
patches.
Original author: Kentoku SHIBA
First reviewer: Jacob Mathew
Second reviewer: Michael Widenius
