MySQL Bugs: #111149: Lock leak in heartbeat_queue

Bug #111149	Lock leak in heartbeat_queue_event()
Submitted:	25 May 2023 9:50	Modified:	8 Aug 2023 0:10
Reporter:	genze wu (OCA)	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: Replication	Severity:	S3 (Non-critical)
Version:	8.0	OS:	Any
Assigned to:		CPU Architecture:	Any

Description:
In sql/rpl_replica.cc:7485, function heartbeat_queue_event() process the heartbeat event with mi->data_lock. But in some error condition, mi->data_lock have not been unlock before return, like in line 7526 and 7536.

After process this error heartbeat event, mi->data_lock leak. All replica command like reset replica will be hang. 

How to repeat:
All error condition can cause that problam, here is an example.

If you have a big transaction（binlog greater than 4GB），After it write to binlog，the binlog will rotate to next file. Set MASTER_HEARTBEAT_PERIOD to a very small value like 0.001, there will be chance that heartbeat event send before binlog rotate.

In this situation, end_pos will overflow and small than log pos on replica point, which will cause this problam.

Suggested fix:
Unlock the mi->data_lock when error found.

affect 8.0

Thank you for your comments

Documented fix as follows in the MySQL 8.2.0 changelog:

    For large transactions （greater than 4GB） and small values
    of MASTER_HEARTBEAT_PERIOD, it was possible for the heartbeat
    event to be sent before binary log rotation could complete,
    causing RESET REPLICA and similar statements on the replica to
    hang.

Closed.

This bug, reported in 8.0, is marked as fixed in 8.2: will this be fixed in 8.0 ?