[Info-vax] Expected votes Was: troubles with quorum disk

[Tom Linden]

On Mon, 07 Jan 2013 09:01:29 -0800, Stephen Hoffman  
<seaohveh at hoffmanlabs.invalid> wrote:

> On 2013-01-07 16:45:35 +0000, Pierre said:
>
>> VAXCLUSTER                      2          1          0          2
>> Coded-value
>> EXPECTED_VOTES                  3          1          1        127
>> Votes
>> VOTES                           1          1          0        127
>> Votes
>> DISK_QUORUM     "$1$DGA4         "    "    "     "    "     "ZZZZ"
>> Ascii
>> QDSKVOTES                       2          1          0        127
>> Votes
>> ...
>>  81518640  TLSAXV  0001000B      3      1     local   member,qf_same
>> 8167FE80  TLSAXY  0001000D      3      1     open    
>> member,qf_same,qf_watcher,qf_active
>> ...
>> I get no error in the autogen, i rebooter the whole cluster but I still  
>> have QDSKVOTES that differ from the quorum disk vote as seen by SDA.
>
> If you have two one-vote voting nodes and a two-vote quorum disk in a  
> two-node cluster, then your EXPECTED_VOTES setting should be four.  Not  
> three.
>
> If you're aiming to survive the loss of one host (as I'd suspect), then  
> set the quorum disk to three votes, and set the EXPECTED_VOTES to five  
> on both hosts.
>
> Or acquire a third (voting) host or emulator for the cluster, and then  
> consider removing the quorum disk entirely, or possibly reducing the  
> QDSKVOTES quorum disk votes to two.
>
> That does mean you'll need the quorum disk around at all times, and  
> you'll encounter the slowdown inherently involved with counting the  
> votes from the quorum disk when either node leaves the cluster.  I'll  
> also presume that Fibre Channel disk is RAID, etc.  But it'll work.
>
> As for the error you're reporting, it's possible that the changes didn't  
> expunge MODPARAMS.DAT of other existing references?  That's pretty easy  
> to do, as older systems tend to find that MODPARAMS file filled with all  
> manner of cruft and duplicate entries.  Check the AUTOGEN summary  
> report(s) in SYS$SYSTEM:AGEN$PARAMS.REPORT, and see what values are  
> listed.  Also confirm that the new (correct) entries are in BOTH files  
> or (if you have a common disk) consider using the AUTOGEN "include file"  
> syntax for the core cluster settings.  And confirm that the AUTOGEN went  
> as far as SETPARAMS or REBOOT, of course.
>
>
Related question.  A while ago we had a power outage that lasted longer  
than the
UPS batteries.  The cluster did not come back up automatically.  The  
licenses had not
been updated on one of the nodes, an oversight.  On closer exam I noted  
that expected_votes
in the 4 node cluster was set at 3 on the other 3 nodes and 7 on the  
errant node.   The cluster
was once 7 nodes but have reduced it.  I tried setting expected votes to 4  
in SYSGEN and did a
WRITE CURRENT  but it did not change it.  What do I need to do?


-- 
PL/I for OpenVMS
www.kednos.com