[Pacemaker] corosync, ocfs2_controld.pcmk insane?
Matthew O'Connor
matt at ecsorl.com
Fri May 11 18:41:38 UTC 2012
On 05/11/2012 12:54 PM, Lars Marowsky-Bree wrote:
> Which of course you can't do if you have actually any OCFS2 file systems
> mounted; that'd result in an immediate suicide of the node.
>
Indeed. :) Happily, I've not created any on this cluster yet. But yes,
that is certainly no solution. I had been working with whatever
versions of the respective packages shipped with Ubuntu 12.04, and on
that test cluster tried configuring and using CMAN. Two things puzzled
me: first, when cman would first crank up on node 1, ocfs2_controld.cman
would immediately start consuming copious CPU (30-50%) until cman was
started on node2. Then it would go quiet. Second, whenever I'd try to
put either node into standby, it seemed like the DLM would fail
catastrophically, taking out Pacemaker with it. It happened almost
every time with that version, and was quite frustrating. With the
resource disabled, both nodes could go in and out of standby without
much, if any, complaint. If this sounds like something worth
investigating also, I would be happy to dig a little deeper.
I'll try strace also next time I see the issue.
Thanks!
-- Matt
More information about the Pacemaker
mailing list