On Wed, May 19, 2010 at 9:25 AM, Gianluca Cecchi <[email protected]> wrote: > On Wed, May 19, 2010 at 8:51 AM, Andrew Beekhof <[email protected]> wrote: > >> On Tue, May 18, 2010 at 2:05 PM, mike <[email protected]> wrote: >> > So now that I have a few clusters up and running after a few problems >> > I've started looking at the logs with some regularity. I'm hoping >> > someone can confirm my thoughts on some entries in the ha-log. >> > >> > 1. PEngine Recheck Timer (I_PE_CALC) just popped! >> > Is this entry related to the next one? >> >> yes >> >> > 2. info: do_state_transition: Starting PEngine Recheck Timer >> > This particular entry appears every 10 minutes or so and I'll see it 3 >> > or 4 times in 1 or 2 minutes and then it will go away for 10 minutes. >> > Then the cycle repeats. >> > >> > 3. info: native_merge_weights: mysql: Rolling back scores from ip_mysql. >> > Now this entry I like. It appears to roll back the fail count (I think) >> >> No. Well not directly. >> This happens when A optionally depends on B and factoring in B's >> allocation preferences would mean A can't run anywhere. >> >> > which is what my DBA was looking for. He wants mysql to failover if >> > there are 3 successive failures of MySQL but only if those successive >> > failures occur within 15 minutes. >> >> You want migration-threshold=3 and failure-timeout=900000 (15 * 60 * 1000) >> >> > Isn't failure-timeout defined as seconds? Or milliseconds?
ms IIRC. Actually you can also specify: failure-timeout=15min http://www.clusterlabs.org/doc/en-US/Pacemaker/1.0/html/Pacemaker_Explained/s-failure-migration.html > Also, in monitor available fields for a resource there are: > > - interval, default 0 > Does it mean no monitor at all if I don't specify a number different from > zero? No. It means non-recurring. http://www.clusterlabs.org/doc/en-US/Pacemaker/1.0/html/Pacemaker_Explained/s-resource-operations.html#s-resource-failure > Is this that determines no monitor at all by default, or is it instead the > "enabled" (see below) field set to false? Or both? > > - enabled, default is true I suppose, if I insert the monitor line inside > the resource, but don't give a value to this field, correct? > > I also found this description about the two parameters migration-threshold > and failure-timeout > > Moving Resources Due to Failure > New in 1.0 is the concept of a migration threshold . Simply define > migration-threshold=N for a resource and it will migrate to a new node after > N failures. There is no threshold defined by default. Todetermine the > resource's current failure status and limits, use crm_mon --failcounts > By default, once the threshold has been reached, node will no longer be > allowed to run the failed resource until the administrator manually resets > the resource's failcount using crm_failcount (after hopefully first fixing > the failure's cause). However it is possible to expire them by setting the > resource's failure-timeout option. > > So, after your comments, suppose migration threshold set to 3 and > failure-timeout set to 15 minutes, is this below the expected behavior? > > time 0 R1 starts > time 5 min R1 fails ---> counter begins for failure-timeout and cluster > successfully restarts it (in place? Where is it set that restart operation > has to be tried in place or on the other node? the fact that failcount < threshold >) ; failcount=1 > time 10 min R1 fails again ---> counter resets for failure-timeout and > cluster successfully restarts it ; failcount=2 > time 25 min R1 has not failed again ---> failcount is reset to 0 and we are > in similar condition as in time 0 its not automatically reset to zero in 1.0 that is new in 1.1 > > Thanks > Gianluca > _______________________________________________ > Linux-HA mailing list > [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems > _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
