[Pacemaker] Very strange behavior on asymmetric cluster
andrew at beekhof.net
Wed Mar 16 15:19:49 EDT 2011
On Wed, Mar 16, 2011 at 6:32 PM, Pavel Levshin <pavel at levshin.spb.ru> wrote:
> 16.03.2011 10:29, Andrew Beekhof:
>>> I still don't understand why it monitors services by default on all nodes
>>> when the cluster is assymetrical
>> Because we don't just make assumptions about what the state of the
>> cluster is - we verify them.
> Actually, you verify what misconfigured or fake resource agents are saying.
RAs are required to be reliable.
Garbage in, garbage out.
> Very sane check, indeed.
> Take DRBD for example. Any resource may be configured on no more than two
> nodes. If there are more nodes in the cluster, N-2 of them will check for
> nothing, fail to check from time to time and cause service instability.
> Imagine a highly overloaded or broken node where all, including fake,
> resource agents are failing to respond in timely manner. Every service in
> such a cluster may suffer from this problem.
> Well, your approach to monitoring is reasonable for symmetrical cluster.
> I beg your pardon if my objections are inappropriate.
> Pavel Levshin
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
More information about the Pacemaker