On 2011-10-08 19:00, [email protected] wrote:
> sry forgot that it is an english list..

It's also a list that doesn't munge the reply-to header. :) I'll assume
you meant to continue the discussion publicly and thus I'm replying to
the list; next time just remember to hit "Reply to All" or "Reply to
List". Else, please clearly mark any personal message to me as "off-list".

> here is my cibadmin
> http://pastebin.com/021Ea2FP

A couple of comments on that one:

That CIB dump reflects a status where both nodes are online, the
resource groups DB and ISCSI aren't running, and their corresponding
DRBD master/slave sets have both instances in the Slave role. That
status is neither expected from the configuration, nor from the
promotion scores that Pacemaker has calculated. Given the CIB status,
one would expect both groups to be running on HA-1, yet they are
reported as being in the Stopped state. Can you guarantee that you did
not produce this CIB dump in the middle of a cluster transition? In
other words, are you sure you waited for the cluster to complete all the
actions it was going to run, before you produced this dump?

I ask because ptest run with your CIB dump as its input does say that
Pacemaker is about to promote drbd-crs, drbd-crs-fail, and drbd-db on
HA-1 next.

In addition, the following location constraints are all redundant:

location SecondaryNode-DB DB 0: HA-2.inta-peters.com
location SecondaryNode-ISCSI ISCSI 0: HA-2.inta-peters.com
location SecondaryNode-OBS OBS 0: HA-1.inta-peters.com
location SecondaryNode-SMB SMB 0: HA-1.inta-peters.com

0 is the placement score that Pacemaker assigns to every node, for every
resource, in a symmetric cluster by default. So you can just remove
those lines from the configuration.

Of the following constraints, the third is redundant:

colocation ISCSI_on_crs inf: ISCSI master-drbd-crs:Master
colocation ISCSI_on_crs_fail inf: ISCSI master-drbd-crs-fail:Master
colocation crs-fail_on_crs inf: master-drbd-crs-fail:Master
master-drbd-crs:Master

There are a few other minor glitches in the configuration, however none
of those (as far as I can see) should prevent the cluster from failing
over properly.

If you're certain that you were _not_ just being impatient, and
Pacemaker does in fact _not_ promote those master/slave sets despite
saying it's going to, then some additional information about your setup
would be helpful. Specifically your distro (CentOS 5.7?) and your DRBD
version.

Cheers,
Florian

-- 
Need help with High Availability?
http://www.hastexo.com/now
_______________________________________________
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