> On Tue, May 13, 2008 at 10:10 AM, Junko IKEDA <[EMAIL PROTECTED]>
wrote:
> > > The resource group is what's not working.  If the second resource in a
> >  > resource group can't start, the first one is stopped, then restarted,
> >  > and you end up with one resource running, one not.  I've attached a
CIB
> >  > that demonstrates this problem, along with a fake RA to simulate the
> >  > resource that can't start (put this in
> /usr/lib/ocf/resource.d/heartbeat).
> >  >
> >  > Here's what I've got, in summary:
> >  >
> >  >    group-test-a (resource group)
> >  >      group-test-a-res-1 (Dummy RA)
> >  >      group-test-a-res-2 (Dummy-that-cant-start RA)
> >  >
> >  >    group-test-b-res-1 (Dummy RA)
> >  >    group-test-b-res-2 (Dummy-that-cant-start RA)
> >  >
> >  >
> >  > group-test-a ends up with group-test-a-res-1 started,
group-test-a-res-2
> >  > stopped.
> >  >
> >  > group-test-b-res-1 and group-test-b-res-2 are bound together with
> >  > colocation and ordering constraints and behave correctly, i.e. they
both
> >  > end up stopped.
> >
> >  Hmm... the same issue?
> >  http://www.gossamer-threads.com/lists/linuxha/dev/43352
> 
> Shouldn't be - he's running a 2.1.3 cluster

I'm not sure but...
2.1.2 would behave as Tim said (not collocation and ordering but group);
> group-test-b-res-1 and group-test-b-res-2 are bound together with
> colocation and ordering constraints and behave correctly, i.e. they both
> end up stopped.

2.1.3 is;
> group-test-a ends up with group-test-a-res-1 started, group-test-a-res-2
> stopped.
 
> >  It seems that if the group's last resource fails,
> >  the other resources would keep starting.
> 
> It might get multiple start events but there shouldn't be any stops.
> The latest version of pacemaker also does away with the redundant start
events.

Year, these changes?
http://hg.clusterlabs.org/pacemaker/stable-0.6/rev/c94066b5d5fe
http://hg.clusterlabs.org/pacemaker/stable-0.6/rev/53f7494c4624

terrific!
It was very hard to explain to customers why some resources start twice.

Thanks,
Junko


_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to