[Pacemaker] WG: configuration of stonith

[Michael Schwartzkopff]

Am Montag, 9. Dezember 2013, 14:58:13 schrieben Sie:
> > > pcs stonith create ipmi-fencing-sv2837 fence_ipmilan
> > 
> > pcmk_host_list="sv2837"
> > 
> > > ipaddr=10.110.28.37 action="off" login=ipmi passwd=abc
> > 
> > delay=15 op monitor
> > 
> > > interval=60s pcs stonith create ipmi-fencing-sv2836 fence_ipmilan
> > > pcmk_host_list="sv2836" ipaddr=10.110.28.36 action="off" login=ipmi
> > > passwd=abc delay=15 op monitor interval=60s
> > > 
> > > pcs property set stonith-enabled=true
> > > 
> > > pcs constraint location ipmi-fencing-sv2837 prefers sv2836=INFINITY
> > > pcs constraint location ipmi-fencing-sv2836 prefers sv2837=INFINITY
> > > 
> > > pcs status
> > > ....
> > > 
> > > Full list of resources:
> > >  ClusterIP	(ocf::heartbeat:IPaddr2):	Started sv2836
> > >  FIXRoute	(ocf::baader:FIXRoute):	Started sv2836
> > 
> > >  ipmi-fencing-sv2837	(stonith:fence_ipmilan):
> > Started sv2836
> > 
> > >  ipmi-fencing-sv2836	(stonith:fence_ipmilan):
> > Started sv2837
> > 
> > This is not optimal. Nothing prevents the resource, that can
> > fence node sv2837
> > to run on host sv2837. You just say, that it should run on
> > node sv2836.
> > 
> > Better would be something like
> > 
> > crm configure location place-fencing-sv2837 -inf: sv2837
> > 
> > or the equivalent in pcs.
> > 
> > Greetings,
> 
> Hi Michael,
> 
> I thought that the lines above will do that:
> > > pcs constraint location ipmi-fencing-sv2837 prefers sv2836=INFINITY
> > > pcs constraint location ipmi-fencing-sv2836 prefers sv2837=INFINITY
> 
> Don't they?

I don't  know pcs in depth and I could not find any detailed doc. So I stick 
with crmsh.

As far as I can judge your lines above your tell the cluster, that the 
resource, that can fence node sv2836, gets INF points if it runs on node 
sv2837. But what happens if node sv2837 is down? Nothing prevents the resource 
starting on the node that it should fence. So if one node is down both fencing 
resources will run on the remaining node. No very nice.

I suggest to assign -INF points to the resource that can fence node sv2836 if 
it runs on node sv2836. So it will run on sv2837. If that node is not 
available the resource cannot run. On the node (remaining) node sv2836 only 
the resource that can fence sv2837.

For details see: http://clusterlabs.org/doc/crm_fencing.html

Mit freundlichen Grüßen,

Michael Schwartzkopff

-- 
[*] sys4 AG

http://sys4.de, +49 (89) 30 90 46 64, +49 (162) 165 0044
Franziskanerstraße 15, 81669 München

Sitz der Gesellschaft: München, Amtsgericht München: HRB 199263
Vorstand: Patrick Ben Koetter, Axel von der Ohe, Marc Schiffbauer
Aufsichtsratsvorsitzender: Florian Kirstein
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 230 bytes
Desc: This is a digitally signed message part.
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20131209/977b0bba/attachment-0004.sig>