Cluster conversion from 3 node to quorum disk

[Aaron Sakovich]

Hi all,

I've got a cluster that used to be a 3 node system -- 2 on a shared SCSI, the third a voting satellite. Things ran great for many, many years.

The satellite has died, and I'm out of hardware and time to replace it. Right now, I'm running on the ragged edge with the 2 remaining nodes up and expected votes set at 3!

What I'm thinking about doing is converting to a (yuk) quorum disk setup, since I've got the shared SCSI bus and an external JBOD shared between the 2 hosts.

Now, I've never ever done a setup with quorum disks before (hard to believe that, but in over 20 years of VMS work I've only done a minimum 3 hosts in a cluster in any config!)

If I'm reading the docu correctly, all I need to do at this time is run Cluster_Config on the 2 nodes to add the quorum disk. Since my expected_votes is ALREADY set to 3, I won't need a reboot of the entire cluster!

But I'm concerned that since Disk_Quorum is not defined (and is not a dynamic parameter), this may not be the case.

Which is correct?

And while I'm talking about this, does anyone have any good tips for tuning items like QDskInterval for this kind of a setup? Is 10 seconds reasonable, or should I set it to something smaller? (Oh, btw, I'm talking about dual DS10 6/617s and a FWD SCSI here.)

Thanks for any insight you might be able to offer!

Aaron

[Volker Halle · Accepted solution]

Aaron,

although this is not something, that I have ever tested, I'm pretty sure you can add the quorum disk to this cluster without a cluster reboot.

Setting DISK_QUORUM="$n$DKAx:" and QDSKVOTES=1 on one of the nodes and rebooting that node should allow the QUROUM.DAT file to be created and used. The other node must be running, so that the cluster has enough votes - even without the votes of the quorum disk. The node will then become a quorum disk watcher. Make sure you do the same to the other node as well, so that both of them can become quorum disk watchers.

The default for QDSKINTERVAL is 3 seconds, so every quorum disk watcher (a node with DISK_QUORUM defined and a direct path to the quorum disk) will issue a read every 3 seconds.

Volker.

[Aaron Sakovich]

I've got to reboot one of the nodes anyways (loading the latest patch set for 7.3-2 this weekend), so that would be cool.

My bad on the 10 seconds -- I'd misread the output from Sysman param show/clu... Three sounds fine to me!

Aaron

[John Gillings]

Aaron,

Best you can ever do is a rolling reboot. All nodes need to agree on the quorum disk, and its votes. DISK_QUORUM is not dynamic. QED.

BUT, please make sure you read the cluster manual relevent to your OpenVMS version and make sure you've configured everything correctly.

Quorum disks on SCSI are even worse than quorum disks anywhere else (if that's possible!). Some versions have had very specific limitations around which disks may be used for quorum, values for QDSKINTERVAL (low value - maybe even 1), use of port allocation classes (mandatory), how to maintain connectivity to the quorum disk during high bus contention (like doing a BACKUP), and maybe some others. Since you haven't stated your version, I can't tell you what applies to you (not that I could without reference to the docs anyway ;-)

[Ian Miller.]

I would not go for a short QDISKINTERVAL as during busy times you can get q disk timeouts.

[Aaron Sakovich]

It's a mixed version cluster, and we're in the planning stage right now to upgrade the lower rev host to match the other. We're running 7.3-2 and 7.1-2 on the 2 nodes. (Let the flames & rants begin... ;^)

We've been running the shared SCSI for several years, so we've already got the alloclass issues ironed out (actually, they were never really a problem -- you should have seen the cluster config we came from! Going to just 2 Alphas with half a dozen disks made the config soooooo much cleaner!)

I'm planning on leaving the QDskInterval at 3 -- that's reasonable for our development environment, I think, even with the 3 being multiplied out due to its granularity.

Thanks!
Aaron

[Jan van den Ende]

Aaron,

first: I agree with Volker.
Although it did also not do that, I have performed the reverse, without any problems.

I also agree VERY MUCH with John:

...even worse than quorum disks anywhere else (if that's possible!).

All that can be said in favor of a quorum disk is, that it IS better than NO quorum supplier.

But old, small systems are readily available quite cheaply (eBay?). (And you already got licenses).
In your config I would NOT choose a VAX, but any small Alpha or Alphastation would do as quorum system!

Of course there are other considerations, and those are yours to solve, but for the money involved, I would just prefer NOT to go quorum DISK, but to use a quorum NODE.

Proost.

Have one on me.

jpe