- Let index_merge allocate table handlers on quick select's MEM_ROOT,
not on statement's MEM_ROOT.
This is crucial for big "range checked for each record" queries, where
index_merge can be created and deleted many times during query exection.
We should not make O(#rows) allocations on statement's MEM_ROOT.
One of them is quite serious: the function table_cond_selectivity used
the TABLE_REF structure for ref/eq_ref access methods as if they had been
filled. In fact these structure are filled after the best execution plan
has been chosen.
The other bugs happened due to:
- an erroneous attempt at get statistics on the result of materialization
of a view
- incorrect handling of ranges with no left/right limits when calculating
selectivity of range conditions on non-indexed columns
- lack of cleanup for some newly introduced fields
Analysis:
The reason for the inefficent plan was that Item_subselect::is_expensive()
didn't detect the special case when a subquery was optimized, but had no
join plan because it either has no table, or its tables have been optimized
away, or the optimizer detected that the result set is empty.
Solution:
Identify the special cases above in the Item_subselect::is_expensive(),
and consider such degenerate subqueries inexpensive.
Add tests crashing the slave in the middle of replication and checking that
replication picks-up again on restart in a crash-safe way.
Fix silly code that causes crash by inserting uninitialised data into a hash.
This bug was introduced by the patch for WL#3220.
If the memory allocated for the tree to store unique elements
to be counted is not big enough to include all of them then
an external file is used to store the elements.
The unique elements are guaranteed not to be nulls. So, when
reading them from the file we don't have to care about the null
flags of the read values. However, we should remove the flag
at the very beginning of the process. If we don't do it and
if the last value written into the record buffer for the field
whose distinct values needs to be counted happens to be null,
then all values read from the file are considered to be nulls
and are not counted in.
The fix does not remove a possible null flag for the read values.
Rather it just counts the values in the same way it was done
before WL #3220.
Test crashing the master, check that it recovers the binlog state.
Fix one bug introduced by previous commit (crash-recoved binlog state was
overwritten by loading stale binlog state file).
Fix Windows build error.
Implement test case rpl_gtid_stop_start.test to test normal stop and restart
of master and slave mysqld servers.
Fix a couple bugs found with the test:
- When InnoDB is disabled (no XA), the binlog state was not read when master
mysqld starts.
- Remove old code that puts a bogus D-S-0 into the initial binlog state, it
is not correct in current design.
- Fix memory leak in gtid_find_binlog_file().
When slave requested to start at some GTID, and that GTID was the very
last event (within its replication domain) in some binlog file, we did
not allow the binlog dump thread on the master to start from the
beginning of a following binlog file. This is a problem, since the
binlog file containing the GTID is likely to be purged if the
replication domain is unused for long.
With this fix, if the Gtid list event at the start of a binlog file
contains exactly the GTID requested by the slave, we allow to start
the binlog dump thread from this file, taking care not to skip any
events from that domain in the file.
Fix MDEV-4329. When user does CHANGE MASTER TO
MASTER_GTID_POS='<explicit GTID state>', we check that this state
does not conflict with the binlog. But the code forgot to give an
error in the case where a domain was completely missing from the
requested position (eg. MASTER_GTID_POS='').
Fix remaining two sporadic test failures in the full test suite:
- Remove crash in DBUG_ASSERT() in _db_flush() (bug also exists in main
tree).
- Fix locking order violation for LOCK_status by temporarily unlocking it
while locking LOCK_active_mi (like a similar fix for
LOCK_global_variables).
Adjust full test suite to work with GTID.
Huge patch, mainly due to having to update .result file for all SHOW BINLOG
EVENTS and mysqlbinlog outputs, where the new GTID events pop up.
Everything was painstakingly checked to be still correct and valid .result
file updates.
During server shutdown, we need to wait for binlog checkpointing to
finish in the binlog background thread before closing the binlog.
This was not done, so we could get assert and failure to finish the
final binlog checkpoint if shutdown happened in the middle.
-Change my_rnd() slightly to make it safer if two threads use it at the same time.
-Avoid some sprintf and strmov in vio.
-Changed thread_count to be automaticly incremented (instead of under LOCK_thread_count).
-Thread cache now uses LOCK_thread_cache instead of LOCK_thread_count.
-Moved delete thd out from LOCK_thread_count.
-Save some mysql_cond_broadcast(&COND_thread_count) calls.
-Removed call to getsockname() during connect.
-Initialize random generator without locks.
Other things:
-Fixed test cases that depends on changes for LOCK_grant
-Added thread_safe_decrement32() and thread_safe_increment32()
-Removed sql_rnd_with_mutex() and get_thread_running()
-In check_table_access() don't lock LOCK_grant if we can resolve the grant with user or db level grants (the normal case).
-Don't use a lock for setting THD->query_id.
-Fixed bug where thd->set_query_id() could be set to same value by multiple threads.
Thanks to Yoshinori Matsunobu for the benchmark of connection speed and to
Domas Mituzas for the inspiration for many of the fixes.
include/violite.h:
Change desc to a string pointer
mysql-test/suite/perfschema/r/all_instances.result:
Added new mutex
mysql-test/suite/perfschema/t/func_mutex.test:
Test for LOCK_system_variables_hash instead of LOCK_grant, as LOCK_grant is not anymore always taken for SELECT's.
mysys/my_gethwaddr.c:
More DBUG
mysys/my_rnd.c:
Change my_rnd() slightly to make it safer if two threads use it at the same time.
sql/event_scheduler.cc:
Changed thread_count to be automically incremented
Moved some safe things out from LOCK_thread_count.
Simplify deleting of THD for running thread.
sql/mysqld.cc:
Changed thread_count to be automically incremented
Thread cache now uses LOCK_thread_cache instead of LOCK_thread_count
Added delete_running_thd()
Moved delete thd out from LOCK_thread_count
More DBUG
Only call mysql_cond_broadcast(&COND_thread_count) if thread_count is 0
Removed call to getsockname() (old not anymore needed check)
sql/mysqld.h:
Removed sql_rnd_with_mutex() (not needed anymore)
Removed not used function get_thread_running()
Added thread_safe_decrement32() and thread_safe_increment32()
Simplified dec_thread_running() and inc_thread_running()
next_query_id() should return the original value for global_query_id, not the next one.
(Bug introduced with MySQL 5.5 merge).
sql/sql_acl.cc:
In check_table_access() don't lock LOCK_grant if we can resolve the grant with user or db level grants (the normal case).
sql/sql_class.cc:
Removed thd_lock_thread_count() and thd_unlock_thread_count()
Initialize random generator without locks
Don't use a lock for setting THD->query_id.
(This is only accessed by thread owning the THD)
sql/sql_class.h:
Don't use a lock for setting THD->query_id.
sql/sql_insert.cc:
Changed thread_count to be automically incremented
sql/sql_parse.cc:
Changed thread_count to be automically incremented
Fixed bug where thd->set_query_id() could be set to same value by multiple threads.
vio/vio.c:
Don't generate 'desc' with sprintf/strmov. Assign a pointer instead.
(Good enough as this is just for debugging)
In some cases, when using views the optimizer incorrectly determined
possible join orders for queries with nested outer and inner joins.
This could lead to invalid execution plans for such queries.
Fix MDEV-4278: Slave does not check that master understands GTID.
Now the slave will abort with a suitable error if an attempt is made to connect
with GTID to a master that does not support GTID.
Fix MDEV-4275 - I/O thread restart duplicates events in the relay log.
The first time we connect to master after CHANGE MASTER or restart, we connect
from the GTID position. But then subsequent reconnects or IO thread restarts
reconnect with the old-style file/offset binlog pos from where it left off at
last disconnect. This is necessary to avoid duplicate events in the relay
logs, as there is nothing that synchronises the SQL thread update of GTID
state (multiple threads in case of multi-source) with IO thread reconnects.
Test cases.
Some small cleanups and fixes.
