[Pacemaker] Migrating VMs that depend on other VMs

Vladislav Bogdanov bubble at hoster-ok.com
Mon May 13 15:19:07 EDT 2013


13.05.2013 19:28, Lindsay Todd wrote:
> Folks: On my cluster built on SL6.4, I need to deploy some VMs that
> depend on other VMs:
> 
>   * db0 has no dependendencies
>   * ldap01,ldap02 require db0 to be running -- ordering constraint, but
>     no collocation (other than we'll use a collocation constraint to
>     recommend ldap01, ldap02 run on different physical hosts)
>   * bq01, bq02 depend on at least one of ldap01, ldap02 running
> 
> At this point, I've got a test arrangement corresponding to only the
> db01 and ldap01 resources being deployed, but it is enough to expose a
> problem...
> 
> The constraints themselves are quite simple.  The trouble arises when we
> try to migrate VMs.  As the "Pacemaker Configuration Explained" manual
> states: "The resource must not, directly or indirectly, depend on any
> primitive or group resources."  What is the reason for this restriction?
> Sure enough, I can't migrate ldap01.  I can migrate db01 successfully,
> though it does force a stop/start on ldap01 (though that is really
> unnecessary, since network connections would have been preserved).
> 
> So is there a way to:
> 
>   * At least get the bq01, bq02 VMs able to migrate, though they depend
>     on other VMs?
>   * Be able to get the ldap01,ldap02 VMs to migrate, without forcing the
>     bq01,bq02 VMs to stop/start (unless both get stopped the restarted)?
>   * Be able to get the db01 resource to migrate, without forcing the
>     other dependents to stop/start, unless it is actually stop/started?

It seems to be exactly what was discussed a year ago here
(http://www.mail-archive.com/pacemaker@oss.clusterlabs.org/msg12050.html)
and then on bugtracker (http://bugs.clusterlabs.org/show_bug.cgi?id=5055).

You may look at whitebox container implementation though, that may fit
your needs. I didn't play with it yet, so that is just a guess.

Alternative way would be to play with tickets as with cluster-wide
attributes, but they are quite limited for that task I think.

Vladislav





More information about the Pacemaker mailing list