On Tue, Sep 21, 2010 at 12:48 PM, Alain.Moulle <[email protected]> wrote:
>>
>> sorry but I don't fully understand :
>> >
>> > ?- I don't think there is any "fencing" functionnality in the OCFS2
>> > management,
>>
>>
>> there is suicide, but that is unrelated to my point
>>
> Ok I didn't not know, but this could not be disturbing for Pacemaker ,
> could it ?

very

> It's quite as if one node crashes for another reason, from Pacemaker
> point of view ...
>>
>>> > ? so for me the membership information remains only at
>>> > Pacemaker/corosync level.
>>>
>>
>> no, ocfs2 also has its own notion of who its peers are - how else does
>> it know who to talk to.
>>
>>
> Yes I know that, it is set in the /etc/ocfs2/cluster.conf, but if one node
> crashes or if Pacemaker decides to fence a node, then the ocfs2 cluster
> knows via its own heartbeat that the peer node is no more alive, again
> I'm sorry but I don't understand where is exactly the problem ... I don't
> mean there is not a problem, but I don't catch it ...

A bunch of filesystem guys and cluster guys got together and thought
through all the corners cases[1] that were possible and decided that
having a resource manager in charge of a cluster FS that was using an
internal membership implementation simply wasn't supportable.

If you want to understand that decision, then you'll need to start
googling the various mailing lists.

[1] Not just when everything works smoothly
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to