[389-users] changelog deadlock replication failures with DNA

Rich Megginson rmeggins at redhat.com
Tue Jun 18 14:03:22 UTC 2013


On 06/18/2013 01:51 AM, thierry bordaz wrote:
> On 06/18/2013 12:35 AM, Mahadevan, Venkat wrote:
>>
>>
>> I do not know why your environment is prone to trigger db_deadlock 
>> (lot of replica agreements, VM, slow disks...).
>> I think the best way to progress is that you fill a ticket/bug so 
>> that we may track the issue. Note this bug is possibly affecting all 
>> operations (ADD/MOD/MODRDN/DEL)
>>
>> -------
>>
>> Hi Thierry,
>>
>> We are running in a VMWare environment, is this known to cause issues 
>> and are there any tuning steps you can recommend or things
>>
>> to watch out for? Thanks and sorry for so many questions.
>>
>> cheers,
>>
>> VM
>>
> Hi Mahadevan,
>
>     Running in a VM should not be an issue if it follows the usual
>     tuning recommendation. But of course it adds a layer when
>     investigating a RC.
>     What is weird is to hit so frequent (50) deadlock in a single DB
>     call. In addition it looks like it always happen when updating the
>     CL so I can think to only RA threads, CL trimming and Write thread
>     (the one holding the backend lock) being in competition.
>
Search threads also. Searches acquire a read lock on the pages used by 
the entry and the indexes.
>
>     You may try to disable the RA for a test to check if they are
>     responsible of that. You may also remove the CL maxage to prevent
>     the CL trimming thread to be in the loop. Also you may check if
>     you have pb with disk IO. Or tune the dbcache to be in memory.
>     Sorry I have no easy answer to your question because it requires
>     deeper investigations.
>
> best regards
> thierry
>
>
> --
> 389 users mailing list
> 389-users at lists.fedoraproject.org
> https://admin.fedoraproject.org/mailman/listinfo/389-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.fedoraproject.org/pipermail/389-users/attachments/20130618/f885c996/attachment.html>


More information about the 389-users mailing list