[ClusterLabs] Antw: Resource getting hard-flipped when node recovers

[Matthew Vernon]

On 15/05/15 08:22, Ulrich Windl wrote:
>>>> Matthew Vernon <mcv21 at cam.ac.uk> schrieb am 14.05.2015 um 19:24 in Nachricht
> <5554DA48.3050709 at cam.ac.uk>:
>> Hi,
>>
>> I have a 3-node pacemaker/corosync cluster (output of crm configure save
>> below); my resource is a xen guest, and if I make its live node standby,
>> then it is live-migrated, as you'd expect.
>>
>> If I hard-power-off the live node, then the other xenhost node starts up
>> the guest, as you'd expect.
>>
>> When, however, I turn the formerly-live node back on, the guest is
>> stopped on the newer-live node, and started on the formerly-live node again.
>>
>> I don't mind this moving, but I do mind that it's a stop-and-start
>> rather than a migration. Can I stop this happening?
> 
> I'd like to know the answer as well (we use a similar scenario), but
> I guess when you "hard powerd off" your node the cluster recorded
> some problem for the node that might not have been cleared when you
> started it up again. Did you check for errors before? (I use to run
> crm_mon -1Arfj)

I didn't, though I was running crm_mon at the time, and don't recall
any. Also, I'm not sure why that should result in a stop-and-start
rather than migrate happening?

Regards,

Matthew