On 8/24/2010 at 12:58 AM, <[email protected]> wrote: > Thank you Tim > It worked but only partially... > Im sorry I feel a bit dumb as I tought it would be automatic.
That's not dumb :) > So I read and added the info that was mentionned in your post. > primitive WMQFS ocf:heartbeat:Filesystem \ > params device="/dev/mapper/360050768019102c24000000000000043_part1" > directory="/var/mqm" fstype="ocfs2" \ > op monitor interval="10s" timeout="10s" on_fail="fence" > OCF_CHECK_LEVEL="20" target-role="Started" > > It worked in the sence that when I deconnect the FC cable it detect the > failure YAY. > But then I'd like it to switch all the ressources over to the other server. > > But it kinda fail because of another resource. > > I run Websphere MQ on this filesystem, so pacemaker try to switch the Queue > Manager to the other server when the drive fail. > At that point it tries to stop the Queue manager but fail, because WMQ just > hang when its filesystem disapear. > (To a point where even as root kill -9 doesn't even work) > So the monitor indicate that the stop failed and it doesn't switch the > resource. > > Should this be fixed by the OCF script (no condition to fail a stop) or by > some other means that would tell > the cluster not to worry about a failed stop? I have zero experience with Websphere, but, in general, if a stop fails the cluster can't migrate the resource to another node, because it might still be running (partially) on the original node. A failed stop should be handled by STONITH, i.e. the node on which the resource failed to stop is killed (rebooted or shut down), provided you have STONITH configured. If there is some actual safe/reliable way to stop Websphere successfully when the filesystem has disappeared out from under it, the Websphere resource agent may need to be enhanced to support this. > I also would like to add a network failure detection, what ocf script can be > used to that, and how to automate the switch > when a failure is detected? > > Thank you all very much. You probably want ocf:pacemaker:ping (this replaces ocf:pacemaker:pingd and ocf:heartbeat:pingd). Theres should be some notes about how to configure this linked off http://www.clusterlabs.org/wiki/Documentation somewhere (there's definitely some description in the Configuration Explained document). Regards, Tim > > > I just finished my first cluster yay. > > > > And Im doing some testing, see how it react to different "stimulus" > > Unfortunately one of my first test failed. > > > > I hope its because I forgot something. > > > > I have 2 nodes > > Node01 owns the ClusterIP, OCFS2 and WebsphereMQ > > > > When I yank out the fiber cable... it doesn't seem to detect it and it > > doesn't switch the ressources on the second node... > > > > What have I forgot? > > I tried adding a op monitor interval=5s to the WMQFS primitive but it > > didn't improve my situation. > You want OCF_CHECK_LEVEL. For further detail see: > http://lists.linux-ha.org/pipermail/linux-ha/2010-March/040009.html > <http://lists.linux-ha.org/pipermail/linux-ha/2010-March/040009.html> > HTH, > Tim > > -- > Tim Serong <[email protected]> > Senior Clustering Engineer, OPS Engineering, Novell Inc. > > > Patrick Ouellet > Administrateur de system Linux > Direction Technologie > Groupe Promutuel > > 418 840-1188, poste 2393 > 1 800 510-4630 > > > > promutuel.ca <http://www.promutuel.ca/> > > Si vous devez imprimer ce document, faites-le recto verso. Si vous n'êtes > pas le destinataire de ce message, veuillez le détruire après avoir informé > l'expéditeur de son erreur. Par ailleurs, il est interdit de copier ou de > modifier tout courriel sans l'autorisation de l'auteur. Promutuel n'assume > aucune responsabilité à l'égard du contenu des messages personnels envoyés > par ses employés. > > > > > > > _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
