On 06/15/2012 11:55 AM, David Vossel wrote:
If resC is stopped
resource stop resC
then drbd_nfsexports is demoted, and resB and resC will stop. Why is
that? I'd expect that resC, being listed last in both the colocation
and
It is the order constraint.
Order constraints are symmetrical. If you say to do these things in this order
1. promote drbd
2. start resB
3. start rscC
Then the opposite is also true. If you want to demote drbd it the following
has to happen first.
1. stop rscC
2. stop resB
3. demote drbd
You can get around this by using the symmetrical option for your order
constraints.
True, but I wasn't demoting DRBD; I was stopping resource C (mostly to
see what would happen if for whatever reason, it couldn't run anywhere).
The order constraint alone doesn't explain the behavior I saw. However,
I think I've pieced together what was actually happening. This constraint:
colocation colo inf: drbd_nfsexports_ms:Master resB resC
means, very confusingly, that to be a master, resB and resC must be
active. Also, for resC to be active, resB must be active. In other words
(with some transitive reduction applied):
drbd -> resC -> resB
It doesn't make any sense, and the documentation is wrong (or at least
self-conflicting). But that's what it really does mean. What was really
confusing is that were it not for the :Master modifier, then crm would
make only one resource set, and it would mean something else. So when I
was testing with dummy resources, I was only more confused. I made
another post explaining it more.
So, DRBD can't be master unless resC and resB are active. And as you
explained, the order constraint will stop resC and resB if DRBD is
demoted. Combined, that means either all these things run, or none:
- resC is stopped, or can't run anywhere
- (by colocation constraint) DRBD may not be master. Demote it.
- (by order constraint) stop resC (actually a no-op: it's already stopped)
- (by order constraint) stop resB
_______________________________________________
Pacemaker mailing list: Pacemaker@oss.clusterlabs.org
http://oss.clusterlabs.org/mailman/listinfo/pacemaker
Project Home: http://www.clusterlabs.org
Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Bugs: http://bugs.clusterlabs.org
