Bug1646100 5.7 #1305
Bug1646100 5.7 #1305
Conversation
| @@ -949,6 +948,8 @@ class MYSQL_BIN_LOG: public TC_LOG | |||
| @retval !=0 Error | |||
| */ | |||
| int get_gtid_executed(Sid_map *sid_map, Gtid_set *gtid_set); | |||
| private: | |||
| void publish_global_status_variables(void) const; | |||
There was a problem hiding this comment.
maybe I would just changed the name of this function to publish_global_snapshot_status_variable.
There was a problem hiding this comment.
Hmm indeed not the best name. publish_position_for_global_status ?
There was a problem hiding this comment.
I think this is better than the previous one. Just note that apart from position you are also publishing binlog_global_snapshot_file. Maybe publish_position_for_global_snapshot_status, but publish_position_for_global_status is good too.
There was a problem hiding this comment.
Ah, "coordinates" then (a pair of name, position), making it publish_coordinates_for_global_status ?
|
I added only one comment. LGTM |
|
@robgolebiowski, please take a look at #1304 too for 5.6 |
In the existing binary log status variables implementation, MYSQL_BIN_LOG::set_status_variables is called with locked LOCK_status and then it attempts to lock LOCK_log. This means that e.g. a long-running commit blocks any SHOW STATUS, and the latter then blocks e.g. client connects/disconnects. While MySQL has not sorted the core server locks by the required locking order, there is no precedent in the core server of LOCK_status owners taking any other mutexes, and most LOCK_status critical sections there are short. This suggests that LOCK_status should come near the end of locking order, and so fix the issue by locking LOCK_log before LOCK_status. For that, create two new static variables binlog_global_snapshot_file and binlog_global_snapshot_position to binlog.cc that hold last "published" global binary log position. Add a new method MYSQL_BIN_LOG::publish_global_status_variables, that, under LOCK_log taken already, locks LOCK_status, and sets the global position variables. Call it in the group commit after flush phase, and optionally after binlog rotate. Change show_binlog_vars to read the global position variables if required (no snapshot; binlog open), and avoid taking LOCK_log there.
In 5.7 this also fixes bug 1657128 (Deadlock by concurrent SHOW BINLOGS, performance_schema.global_status query, and binlog purge). The merge adds the testcase and its required PFS debug sync point.
laurynas-biveinis
force-pushed
the
bug1646100-5.7
branch
from
873e734
to
9a4a7a3
Compare
|
Updated method name |
|
LGTM |
http://jenkins.percona.com/job/mysql-5.7-param/584/
@robgolebiowski, please review as well