[Rdo-list] [Mitaka] Automatic restart for services

Haïkel hguemar at fedoraproject.org
Thu Nov 12 14:05:27 UTC 2015


2015-11-12 14:30 GMT+01:00 Perry Myers <pmyers at redhat.com>:
> On 11/12/2015 08:06 AM, Alan Pevec wrote:
>> 2015-11-12 14:00 GMT+01:00 Perry Myers <pmyers at redhat.com>:
>>> can you describe what you mean here by automatic restart for services?
>>>
>>> I'm worried what the implications would be for things that are running
>>> under Pacemaker control in an HA environment, so I think we need more
>>> details here
>>
>> Can we have more details what "under Pacemaker control" means?
>> Haikel's proposal is about modifying systemd service unit files but
>> afaik Pacemaker is not using them but what they call "resource agents"
>> ?
>
> Not correct... not every service _has_ a resource agent. Mariadb/galera
> and rabbitmq do, but all other services are controlled via Pacemaker via
> their systemd scripts
>
> So when you tell pacemaker to stop nova-api, it does so via systemd
>

This is ok, if you stop normally services, they won't be restarted.


> If something outside of pacemaker knowledge (like user from command
> line, or rpm %post or whatever) uses systemd directly to stop nova-api,
> pacemaker says "OMG the service died" and tries to recover
>
> Perry
>

If Andrew (or anyone else) could confirm, but from my understanding,
Pacemaker already disable systemd auto-respawn since 1.1.10 (which is
quite old, RHEL7 ship 1.1.12 at least)
https://github.com/ClusterLabs/pacemaker/commit/37c6efae3b9d1db18d8d782bd8ff1da7b77f14f3
https://github.com/ClusterLabs/pacemaker/blob/master/lib/services/systemd.c#L564
<= the code generating the override file shows that it does disable
that feature/

Concerning the changes, they will considerably improve reliability for
people not relying on Pacemaker, so I guess we're good in that aspect.

Regards,
H.




More information about the dev mailing list