[JIRA] Commented: (SSPCPP-473) occasional shibd crashes when using MySQL ODBC storage service on RHEL6

[BRIAN P KROTH (JIRA)]

    [ https://issues.shibboleth.net/jira/browse/SSPCPP-473?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14487#comment-14487 ] 

BRIAN P KROTH commented on SSPCPP-473:
--------------------------------------

FYI, this is biting me too on Debian 6.

I've found that if I backport libmyodbc to 5.1.10 it fixes the pthread_mutex crashes that can happen with high concurrency (eg: ~30 simultaneous connections), but still has a pretty huge memory leak (on a busy site the shibd process will grow to over 2G by the end of the day - ~50M/2000 requests).  I'm still trying to track down the memory leak.

Also, backporting to shibd 2.5.0 didn't help.

Thanks,
Brian

> occasional shibd crashes when using MySQL ODBC storage service on RHEL6
> -----------------------------------------------------------------------
>
>                 Key: SSPCPP-473
>                 URL: https://issues.shibboleth.net/jira/browse/SSPCPP-473
>             Project: Shibboleth SP - C++
>          Issue Type: Bug
>      Security Level: Standard(Standard bug, may impact functionality but does not represent a security vulnerability ) 
>    Affects Versions: 2.4.3
>         Environment: RHEL6, Apache 2.2, SP 2.4.3 from RPM, and MySQL for storage service. Using RH supplied unixODBC and mysql-connector-odbc packages with default odbcinst.ini.
>            Reporter: Sebastien Otto Korner
>            Assignee: Scott Cantor
>         Attachments: shibd-crash-mode-1.txt, shibd-crash-mode-2.txt
>
>
> When using the ODBC storage service with MySQL on RHEL6 shibd crashes periodically.  Our web servers are 1/2 RHEL5 and 1/2 RHEL6, with load spread equally between them.  The shibd on RHEL5 never crashes.  Without using the ODBC storage service shibd never crashes on any server.
> The crash frequency in terms of session creation is about once every thousand.  It does not appear to be a regular occurrence based on number of sessions since the last crash though.  For approximately half the crashes another crash follows within 30 minutes.
> Investigating the core files indicates two different failure modes, with MySQL use implicated in both.  Attached are traces representing the two failure modes.
> Let me know if you require more information.
> Thanks.
> Sebastien

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira