Hi By the way, is there a description somewhere of parameters from /etc/sysconfig/pacemaker ?
Thanks Alain De : Dejan Muhamedagic <deja...@fastmail.fm> A : General Linux-HA mailing list <linux-ha@lists.linux-ha.org> Date : 21/11/2011 15:48 Objet : Re: [Linux-HA] Antw: What about "start-delay" attribute status ? Envoyé par : linux-ha-boun...@lists.linux-ha.org On Mon, Nov 21, 2011 at 03:07:43PM +0100, alain.mou...@bull.net wrote: > Thanks Dejan, > ok I understand, so we have to choose between a small value of > LRMD_MAX_CHILDREN > and on start, stop, or status of 64 resources it will take a while ... > and a big value of LRMD_MAX_CHILDREN and then either the start, stop and > at best, status will be achieved very quickly as they are parallelized or > at > worst the system will be "on knees" ... > We'll give it a try ... as I have big computers ;-) Just note that you should try to think of every possible combination of resource operations. For instance, imagine 64 Xen VMs trying to start in parallel. Better be conservative than to push your nodes to their limit. > But my question is now : when you write : > "Let me just add that operations which were supposed to > start at the same time get spaced out." > So if LRMD_MAX_CHILDREN=4, that means that if ask for start on 32 > resources at the > same time, Pacemaker will mange 4, delay the remaing 28, manage 4 again, > etc. so > it will be completed in 8 shots, right ? No. > But what is the delay value between each shot ? There is none. As soon as one operation finishes, another one gets started. Now, if you have say four big RDBMS instances starting and each of them takes five minutes or so, the other resources will obviously stay in the queue for five minutes. Anyway, you can see for yourself on cluster start, just grep your logs for lrmd:.*rsc:, it should show you all timestamps when certain operation was started (apart from recurring monitors). Thanks, Dejan > Thanks > Alain > > > > > De : Dejan Muhamedagic <deja...@fastmail.fm> > A : General Linux-HA mailing list <linux-ha@lists.linux-ha.org> > Date : 21/11/2011 13:45 > Objet : Re: [Linux-HA] Antw: What about "start-delay" attribute status ? > Envoyé par : linux-ha-boun...@lists.linux-ha.org > > > > Hi, > > On Mon, Nov 21, 2011 at 01:42:15PM +0100, alain.mou...@bull.net wrote: > > Hi Florian, > > ok I've checked the thread, so that means that on RHEL6 , if I have > let's > > say 32 resources groups of 2 primitives on > > each node, I can set the LRMD_MAX_CHILDREN environment variable in > > /etc/sysconfig/pacemaker to 64 ? > > The number of resources shouldn't be the main criteria for > setting this parameter, but what can your nodes handle without > being overloaded. So, 64 sounds sounds like you have some really > big computers :) It also depends on the nature of the cluster > resources. The default of 4 is rather conservative, perhaps > nowadays 8 would be better. > > > Is it acceptable for lrmd and Pacemaker ? Or will we face any > side-effect > > ? > > LRMD_MAX_CHILDREN is the maximum number of resource operations > allowed to run in parallel. Hope that that answers your question. > > Thanks, > > Dejan > > > Thanks > > Alain > > > > > > > > De : Florian Haas <flor...@hastexo.com> > > A : General Linux-HA mailing list <linux-ha@lists.linux-ha.org> > > Date : 21/11/2011 12:58 > > Objet : Re: [Linux-HA] Antw: What about "start-delay" attribute status > ? > > Envoyé par : linux-ha-boun...@lists.linux-ha.org > > > > > > > > On 11/21/11 13:03, alain.mou...@bull.net wrote: > > > Hi, > > > yes that's exactly the purpose of my question (and exactly the same > > > problem of "big-monitoring-trains") : > > > if we can always use start-delay to ramdomize the first monitor > > operation > > > time on all the resources on a server, > > > but if it is really deprecated, that means that in the future this > > option > > > will no more > > > be managed by Pacemaker (perhaps it already is the case ... ?) , so in > > > > this case > > > we must not use this option. > > > > > > Could someone give us a clear status on this option "start-delay" ? > > > > If your RA needs it, then the RA is most likely broken. :) > > > > For monitor operations allegedly piling up, please consider this: > > http://www.gossamer-threads.com/lists/linuxha/pacemaker/76152#76152 > > > > Hope this helps. > > Cheers, > > Florian > > > > -- > > Need help with High Availability? > > http://www.hastexo.com/now > > _______________________________________________ > > Linux-HA mailing list > > Linux-HA@lists.linux-ha.org > > http://lists.linux-ha.org/mailman/listinfo/linux-ha > > See also: http://linux-ha.org/ReportingProblems > > > > _______________________________________________ > > Linux-HA mailing list > > Linux-HA@lists.linux-ha.org > > http://lists.linux-ha.org/mailman/listinfo/linux-ha > > See also: http://linux-ha.org/ReportingProblems > _______________________________________________ > Linux-HA mailing list > Linux-HA@lists.linux-ha.org > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems > > _______________________________________________ > Linux-HA mailing list > Linux-HA@lists.linux-ha.org > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems _______________________________________________ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems _______________________________________________ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
