On Fri, Dec 14, 2012 at 7:53 PM, Emmanuel Saint-Joanis
<[email protected]> wrote:
>> Assuming the second node does not run pacemaker .... switch the cluster
>> into maintenance-mode, adjust your crm configuration to reflect your
>> changes to the VMs and restart pacemaker.
>>
>> Once you made sure all is in place and all vm configurations are
>> successfully probed disable maintenance-mode and start pacemaker on the
>> second node.
>
>
> I thank you Andreas for those enlightenments, and I'd like to submit my
> case which is quite related :
> I bumped sometimes in cases where the cluster was stuck, resources not
> willing to start, so I did :
> crm configure save /tmp/backup
> stop pacemaker & corosync
> rm /var/lib/heartbeat/*/* /var/lib/pengine/*
> start corosync & pacemaker
> crm configure load update /tmp/backup

Thats basically the same as:

stop corosync/pacemaker
start corosync/pacemaker

The resource state is discarded when the cluster stops.
But do please create a bugreport the next time this happens, if
pacemaker is misbehaving we need to know about it so we can fix it.

>
> Those actions seemed to fully unlock all the nodes starting up happily -
> but I have some dislike for this ultimate weapon.
>
> Then, do you know a way to ->reset<- all states ? I mean what we see by
> doing cibadmin -Q, which represents current states.

cibadmin --erase
or crm_resource --reprobe

Check the man pages for what they do.

> By the way, is there some other kind-of memory of cluster-history, such as
> current scores of resources, for example ?

The policy engine files can be replayed to show this.
_______________________________________________
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