[ClusterLabs] Trying to Understanding crm-fence-peer.sh

[Andrei Borzenkov]

16.01.2019 19:49, Bryan K. Walton пишет:
> On Wed, Jan 16, 2019 at 04:53:32PM +0100, Lars Ellenberg wrote:
>>
>> To clarify: crm-fence-peer.sh is an *example implementation*
>> (even though an elaborate one) of a DRBD fencing policy handler,
>> which uses pacemaker location constraints on the Master role
>> if DRBD is not sure about the up-to-date-ness of that instance,
>> to ban nodes from taking over the Master role.
>>
>> It does NOT trigger node level fencing.
>> But it has to wait for, and rely on, pacemaker node level fencing.
> 
> Thanks Lars.  Between these comments, and the man page for drbd.conf, I
> think I understand what is going on here.  Is it correct to say, that in
> the case I provided, that DRBD successfully issued a "drbdadm outdate
> res" on the other node, and therefore it didn't need to STONITH the
> peer?

That is not what your logs show. DRBD successfully issued fence-peer
handler on local node. crm-fence-peer.sh won't attempt to run anything
on other node. DRBD just calls fence-peer handler, it is up to handler
to issue explicit stonith request if appropriate.

>  (Looking at the crm-fence-peer code, I see that exit code 4 is
> node fenced, but there is an exit code 7 which means STONITHED.  In my
> case, I got an exit code 4, and not a 7.)
> 

crm-fence-peer.sh return 7 when it detected that pacemaker performed
STONITH on peer on its own. crm-fence-peer.sh does not initiate stonith
by itself (as far as I can tell).

> Also you mentioned that "Other implementations of drbd fencing policy
> handlers may directly escalate to node level fencing."
> 
> Are these "other implementations" third-party handlers?  Or are they
> available from within the DRBD software? Can you point to any of these?
> 

stonith_admin-fence-peer.sh?