[ClusterLabs] Antw: [EXT] why is node fenced ?

[Andrei Borzenkov]

30.07.2020 23:23, Lentes, Bernd пишет:
> 
> 
> ----- Am 30. Jul 2020 um 9:28 schrieb Ulrich Windl Ulrich.Windl at rz.uni-regensburg.de:
> 
>>>>> "Lentes, Bernd" <bernd.lentes at helmholtz-muenchen.de> schrieb am 29.07.2020
>> um
>> 17:26 in Nachricht
>> <1894379294.27456141.1596036406000.JavaMail.zimbra at helmholtz-muenchen.de>:
>>> Hi,
>>>
>>> a few days ago one of my nodes was fenced and i don't know why, which is
>>> something i really don't like.
>>> What i did:
>>> I put one node (ha-idg-1) in standby. The resources on it (most of all
>>> virtual domains) were migrated to ha-idg-2,
>>> except one domain (vm_nextcloud). On ha-idg-2 a mountpoint was missing the
>>> xml of the domain points to.
>>> Then the cluster tries to start vm_nextcloud on ha-idg-2 which of course
>>> also failed.
>>> Then ha-idg-1 was fenced.
>>
>> My guess is that ha-idg-1 was fenced because a failed migration from ha-idg-2
>> is treated like a stop failure on ha-idg-2. Stop failures cause fencing. You
>> should have tested your resource before going productive.
> 
> Migration failed at 16:59:34.
> Node is fenced at 17:05:35. 6 minutes later.
> The cluster needs 6 minutes to decide to fence the node ?
> I don't believe that the failed migration is the cause for the fencing.
> 


With default settings node is fenced if

a) node is unreachable
b) stopping resource on this node failed