[Info-vax] Quorum disk and last transition time

[Stephen Hoffman]

On 2015-01-28 15:49:31 +0000, Joe said:

> We have a quorum disk configured on 2 node cluster. If one of the 
> quorum disk watchers loses this disk momentarily, will a cluster state 
> transition be triggered?

If both voting hosts are present and accounted for, and both have one 
vote each, and the quorum disk has one vote and there are three votes 
total, no.   The quorum disk votes "matter" when there's not enough 
non-quorum-disk votes present to meet quorum.  When the quorum disk 
vote matters, there'll be pause of 3 x the polling interval, to ensure 
there are no unrecognized hosts sharing access to the quorum disk.

> we receive these messages in OPCOM when the disk goes out.

If that's SCSI, then that's probably a flaky and failing quorum disk, 
or a flaky and failing device somewhere else on the (presumably shared) 
bus, maybe bus termination or a controller problem, or a bad or loose 
cable.  If that's a Fibre Channel or DSSI bus involved, then adjust the 
details accordingly.  In any case, investigate the devices and the 
error logs for details, and expect to have to swap something.

> Could the above event trigger a state transition and where can I find 
> the last transition time due to the above event?

See above.  As for the logging, the system consoles are typically the 
only "working" output devices during a quorum transition, and what 
happens with the consoles is limited to output messages.   OPCOM and 
other processes are intentionally wedged.  Any user-mode software and 
any of the lower-priority kernel-mode VMS code is not going to be 
stalled, pending the detection of quorum and the resumption of 
processing.  That stall includes the ability to write to disk; there 
won't be logging or other activity happening until after the stall is 
cleared.

> If we give $ show cluster/cont and add cluster, the field 
> LAST_TRANSITION always show the last transition time of one of the 
> nodes which ever is later, will this field be updated if a cluster 
> transition is triggered due to quorum disk validation/invalidation.

I'd not be interested in the cluster state transitions, and would be 
far more centrally interested in what's up with that quorum disk and 
whatever bus is involved.

-- 
Pure Personal Opinion | HoffmanLabs LLC