Percona server 5.5 completely stalls during import (myloader)

Bug #1519094 reported by Mischa ter Smitten
8
This bug affects 1 person
Affects Status Importance Assigned to Milestone
MySQL Server
Unknown
Unknown
Percona Server moved to https://jira.percona.com/projects/PS
Fix Released
High
Laurynas Biveinis
5.5
Fix Released
High
Laurynas Biveinis
5.6
Fix Released
High
Laurynas Biveinis

Bug Description

http://dba.stackexchange.com/questions/120923/mysql-percona-server-5-5-completely-stalls-during-import-myloader

The problem seem to be caused by: innodb_adaptive_flushing_method = keep_average, innodb_adaptive_flushing_method = estimate runs stable.

description: updated
Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

Output of:

gdb -ex "set pagination 0" -ex "thread apply all bt" --batch -p $(pidof mysqld)

Revision history for this message
Laurynas Biveinis (laurynas-biveinis) wrote :

Unfortunately some of the symbol information is missing in gdb dump, does the server have debug info packages installed?

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

No, how does that work? Installing the percona-server-5.5-dbg package and then what?

Revision history for this message
Laurynas Biveinis (laurynas-biveinis) wrote :

Please just retry the gdb command after having the package installed.

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

I will do that.

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

Output of:

gdb -ex "set pagination 0" -ex "thread apply all bt" --batch -p $(pidof mysqld)

With percona-server-5.5-dbg installed

Revision history for this message
Laurynas Biveinis (laurynas-biveinis) wrote :

Thanks. I am analysing the dump.

Did this started happening after an upgrade? If yes, can you tell the latest PS 5.5 version which worked for you?

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

Hard to say, it's on new hardware, new installation.

On our previous setup we never encountered the problem (in our normal workload). But we never do mass imports there and we also have different hardware there so it's hard to compare...

tags: added: i62298
Revision history for this message
Laurynas Biveinis (laurynas-biveinis) wrote :

Mischa and everybody experiencing random hangs on 5.5.46 / 5.6.27, please test server build with a preliminary fix, available at:

5.5: http://jenkins.percona.com/view/Percona-RELEASES/job/percona-server-5.5-RELEASE/148/
5.6: http://jenkins.percona.com/view/Percona-RELEASES/job/percona-server-5.6-RELEASE/50/

(please click through the blue dots to navigate to the right packages for your system).

Please confirm whether with these packages the problem does not occur anymore. In case it still does, please repeat the previous procedure: 1) make sure the debug info package is installed; 2) get the server to hanged state; 3) execute "gdb -ex "set pagination 0" -ex "thread apply all bt" --batch -p $(pidof mysqld)" and post here its output.

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

I'll try to test them today.

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

Is there any reason why I would need to avoid using innodb_adaptive_flushing_method = keep_average on non SSD cards.

On our machines we have:
* single SSD
* multiple SSD, software RAID
* multiple SSD, hardware RAID

But non of them have SSD cards.

Revision history for this message
Laurynas Biveinis (laurynas-biveinis) wrote :

I assume this is asked in the context of the current bug, i.e. why keep_average causes this hang. But it's not keep_average itself which is the cause of the bug (if the fix is confirmed), but rather increased mutex pressure. The bug has probability of appearing with any adaptive flushing setting.

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

Yes, in the context of the current bug.

We don't have the problem when we use innodb_adaptive_flushing_method = estimate and when using innodb_adaptive_flushing_method = keep_average on MariaDB.

Revision history for this message
Laurynas Biveinis (laurynas-biveinis) wrote :

If our analysis is correct then 1) the bug has a probability of appearing with any innodb_adaptive_flushing_method on Percona Server, just some more likely than others 2) MariaDB is not affected with any innodb_adaptive_flushing_method.

We will post the final bug analysis after we have confirmations that the issue disappeared.

Revision history for this message
Mischa ter Smitten (mischa-ter-smitten) wrote :

I've had 52 successful imports (before patch not more than +/- 5) last night so it looks like it fixed our problem.

tags: added: innodb upstream
Revision history for this message
Laurynas Biveinis (laurynas-biveinis) wrote :
Revision history for this message
Yura Sorokin (yura-sorokin) wrote :

Fixed by Oracle in 5.5.49, 5.6.30 and 5.7.12.

5.5 commit: 93e6f38
https://github.com/mysql/mysql-server/commit/93e6f388860490a6066cc07253a6aab94029e502

Revision history for this message
Shahriyar Rzayev (rzayev-sehriyar) wrote :

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-945

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.