[Pacemaker] Pcmk migration logic and Libvirt migration behavior

David Vossel dvossel at redhat.com
Thu May 2 17:41:16 UTC 2013


----- Original Message -----
> From: "Andreas Hofmeister" <andi at collax.com>
> To: pacemaker at oss.clusterlabs.org
> Sent: Thursday, May 2, 2013 9:32:36 AM
> Subject: Re: [Pacemaker] Pcmk migration logic and Libvirt migration behavior
> 
> On 05/01/2013 10:49 PM, David Vossel wrote:
> 
> >
> > Have you tested this with 1.1? There have been changes to how migration
> > works, some of which have to do with properly handling partial migrations.
> > I'd test this in 1.1.10.rc1 or anything >= 1.1.8
> >
> 
> No, we did not try with 1.1 yet.
> 
> The patches you mentioned seem to fix the problem where Pcmk did not
> call "migrate_from" after a "migrate_to" ? I my case, this is not(much)
> of a problem since here, "migrate_from" is basically a no-op anyways.
> 
> The problem I described essentially boils down to the fact that Pcmk
> assumes a resource to be stopped after a successful "migrate_to". But
> the basic assumptions on how a migration work did not seem to have
> changed that much in 1.1 (?)

I see... This is kind of complicated.

If a migration is interrupted, I don't think it is safe for pacemaker to have any implied assumptions about the state of the resource.  We might need to have the pengine issue a re-probe in this situation.

Any chance you can share your pengine transition files?  Making a crm_report for the time frame the issue occurred in would gather this information for you.

-- Vossel

> 
> 
> Ciao
>    Andi
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
> 




More information about the Pacemaker mailing list