[Pacemaker] DRBD Master/Slave in a 3 node cluster

[James Oakley]

On Tuesday, October 1, 2013, Stefan Botter <listreader at jsj.dyndns.org> wrote:
> I have a quite similar setup, currently running on stock 12.2. I have
> a test
> system just updated to 12.3, with the ha-clustering:Stable,  and it
> fails
> with STONITH enabled almost instantly, due to certain segfaults in
> the
> stonith resources.
> With 12.2 it works flawless, and with 12.3 and the Stable repo, but
> without STONITH, also.

I haven't tested STONITH on this particular cluster yet, since I need to get DRBD running reliably first. STONITH at this point will just immediately guarantee split-brain, rather than prevent it.


> I don't know, but my configuration is - as said - almost similar, but
> a
> _lot_ shorter, due to usage of groups and thus far less contraints
> and
> location definitions. My nodes are virtual machines in VMware, thus
> the
> vcenter stonith resources. The nodes, hermes1 and hermes 2 have the
> drbd resources, hermes1 being the preferred node, and hermes3 is
> there for quorum (and logs):

It seems that drbd slaves are working OK for you. My problem is that Pacemaker will only run the master part. It will happily attempt to run the master on either of the DRBD nodes (and the non-DRBD node...) but it flat out refuses to run the slave. It doesn't even try, and it will actually shut down the secondary DRBD configuration if I bring it up manually. I can't find a single log message to indicate why, and I can't find a command or configuration option to say "run this slave on this node."


> As said, this setup currently runs on openSUSE 12.2
> I know, 13.1 is near, but I fear the status of the ha-clustering in
> 13.1 will not be that great, so maybe you give it a try with a 12.2
> installation first.

I'm using some stuff only available in 12.3, and openSUSE won't downgrade cleanly. I'd rather just get the issue fixed.

-- 
James Oakley
jfunk at funktronics.ca