On Thursday 12 November 2009, Andrew Beekhof wrote: > On Thu, Nov 12, 2009 at 11:54 AM, Bernd Schubert > > <bs_li...@aakef.fastmail.fm> wrote: > > Hello, > > > > I try to prevent auto-migration back from mds2 to mds1, but somehow > > resource- stickiness doesn't seem to work. After a failure of mds1 and > > takeover on mds2, it still migrates back to mds1 when this system comes > > back > > > > primitive MDT_HC3WORK ocf:ddn:lustre_server \ > > params device="/dev/vg_HC3WORK/mdt" > > directory="/lustre/HC3WORK/mdt" \ op start interval="0" timeout="700" \ > > op stop interval="0" timeout="600" \ > > op monitor interval="120" timeout="600" \ > > meta resource-stickiness="200" is-managed="true" > > > > > > property $id="cib-bootstrap-options" \ > > default-resource-stickiness="200" \ > > no-quorum-policy="stop" \ > > dc-version="1.0.6-f709c638237cdff7556cb6ab615f32826c0f8c06" \ > > cluster-infrastructure="Heartbeat" > > > > > > location location-MDT_HC3WORK.mds1 MDT_HC3WORK 100: mds1 > > location location-MDT_HC3WORK.mds2 MDT_HC3WORK 50: mds2 > > location location-MDT_HC3WORK.oss1 MDT_HC3WORK -inf: oss1 > > location location-MDT_HC3WORK.oss2 MDT_HC3WORK -inf: oss2 > > location location-MDT_HC3WORK.oss3 MDT_HC3WORK -inf: oss3 > > location location-MDT_HC3WORK.oss4 MDT_HC3WORK -inf: oss4 > > > > > > MDT-HC3WORK is also part of a resource group, but the resource group and > > also all member do have the very same location constraints > > that matters > > >. Is is my mistake or a > > bug? > > check the current score with ptest -Ls and make sure the stickiness it > higher. or just set it to INFINITY to make sure it never moves.
Ah, thanks! I didn't know about this command. It lists all resources several times, those that are part of a resource group even with different numbers. Up to 600, no surprise it migrates the resource. I thought I only have to set it higher than the location constraint. Hmm, any chance you could rename "ptest" to "crm_allocation_score" or so? Associating from crm or pacemaker to "ptest" is rather difficult and so the command is not easy to remember. Thanks a lot again, Bernd -- Bernd Schubert DataDirect Networks _______________________________________________ Pacemaker mailing list Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker
