On 12/07/13 00:59, Andrey Groshev wrote:
11.07.2013, 18:48, "Digimer" <li...@alteeve.ca>:
You need fencing. Specifically, cman blocks when a fence is called and
won't unblock until it's told that a fence completed successfully.
Configure cluster.conf to use 'fence_pcmk', which tells cman to pass
fence requests to pacemaker, and then configure (and test!) stonith in
pacemaker.
If you have just two nodes, be sure to also set '<cman two_node="1"
expected_votes="1" />'.
digime
On 11/07/13 09:35, Andrey Groshev wrote:
I understand that it may be correct to do so...
But why so difficult?
Assume, I make a small НА cluster in my garage.
And, I not have a managed switch or managed UPS.
I can not corrupt the data.
I just need to returning node as soon as possible started responding.
First, fencing is not difficult, it's just one of the parts of
clustering to learn. Second, the cluster software has no concept of
"unimportant cluster". It treats every cluster as enterprise class.
Third, if a service can run on both nodes without coordination with one
another, then you don't need the cluster stack at all.
Assuming you actually need to keep the service on one node or the other,
you need to have a way to make sure the "lost" node really is lost. You
are not allowed to make assumptions. "The only thing you know if what
you don't know". Fencing puts a node in an unknown state (disconnected?
frozen? crashed? blown to pieces?) and puts it into a known state,
"off". This ensures the service can never run on both nodes at the same
time.
--
Digimer
Papers and Projects: https://alteeve.ca/w/
What if the cure for cancer is trapped in the mind of a person without
access to education?
_______________________________________________
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
