Bug #67384 Increasing memory consumption

• Submitted: 26 Oct 2012 10:04
• Modified: 19 Jun 2013 8:44
• Reporter: Thomas Schulz
• Status: Duplicate
• Category: MySQL Server: Replication
• Severity: S2 (Serious)
• Version: 5.1.66
• OS: Linux (openSUSE 11.4 (x86_64))
• CPU Architecture: Any

[26 Oct 2012 10:04] Thomas Schulz

Description:
After updating mysql server from 5.1.63 to 5.1.65/5.1.66, the system shows an permanent increasing consumption of main memory after 2-3 days (more than before). With 5.1.63 the memory consumption raises only to a certain level (60% of 2 GB) and remains there.

The server is running as slave from live system with approximately 30 inserts/updates/replaces per second. MySQL is installed as "Generic Linux TAR Archive".

How to repeat:
Going back to 5.1.63 lets disappear the effect and updating to 5.1.65 ot 5.1.66 lets occur it again.

[26 Oct 2012 11:00] MySQL Verification Team

Hi Thomas, this is an interesting observation.  I think number of changes made to the 5.1 tree is not that many, so I'll try manually checking what changes might have led to a memory leak.

Can you upload the my.cnf for this instance..  Also, let us know:

o) does "FLUSH TABLES;STOP SLAVE;START SLAVE;"  reduce the memory at all?
o) using innodb plugin or builtin innodb?
o) got any queries that contain @user variables in them?
o) does disabling query cache fix anything?

[26 Oct 2012 11:00] Valeriy Kravchuk

As this is a slave, do you have any replicate* options set?

[26 Oct 2012 12:15] Thomas Schulz

Can you upload the my.cnf for this instance..  
=> done

o) does "FLUSH TABLES;STOP SLAVE;START SLAVE;"  reduce the memory at all?
=> test follows
o) using innodb plugin or builtin innodb?
=> builtin
o) got any queries that contain @user variables in them?
=> as far as I know - no
o) does disabling query cache fix anything?
=> test follows

[26 Oct 2012 12:16] Thomas Schulz

As this is a slave, do you have any replicate* options set?

yes:
replicate-ignore-table = mysql.db
replicate-ignore-table = mysql.user

[26 Oct 2012 12:52] Valeriy Kravchuk

See https://bugs.launchpad.net/percona-server/+bug/1042946 also. Many users suspect that in recent versions we have a leak of a kind when replicate* options are used on slave.

[29 Oct 2012 12:54] Thomas Schulz

After this weekend I can also answer the following questions:
o) does "FLUSH TABLES;STOP SLAVE;START SLAVE;"  reduce the memory at all?
=> no
...
o) does disabling query cache fix anything?
=> no

[22 Dec 2012 21:30] MySQL Verification Team

although i cant repeat this bug, it appears obvious to me that the cause is a regression introduced in the fix for http://bugs.mysql.com/bug.php?id=54201

[3 Jan 2013 8:12] MySQL Verification Team

Marking as duplicate of bug #67983 since that one has shorter description and clear cause.

[19 Jun 2013 8:44] Thomas Schulz

Fixed in 5.1.70.