Question regarding system files in a 7.3-2/8.3 mixed version cluster

[Dana Conroy]

I have a four-node, all-Alpha cluster which is split across two sites. Site A and Site B each have two nodes - Node1 and Node3 are at Site A; Node2 and Node4 are at Site B. Node1 is the production Alpha, Node2 is the DR/test Alpha which has a test application/database environment, and Node3 and Node4 are small Alphas that provide votes and help keep shadow members available to the cluster. The cluster has two system disks, one at each site. All nodes are currently running OpenVMS 7.3-2.

I am planning to perform a rolling upgrade from 7.3-2 to 8.3 where I (1) upgrade the site B nodes to 8.3, (2) run a mixed-version cluster for a few weeks while testing the database and application environment on Node2, then (3) update the site A nodes to 8.3.

Referring to Section 4.5.6 ("Returning Authorization and AGEN$INCLUDE Files to the System Disk", pages 96-97) in the OpenVMS 8.3 Upgrade and Installation Manual, it states on page 96:

* "...If any logical name points to a location or file name other than the location and file name listed in Table 4-2, return the file to the default location and file name..."

* "...After the upgrade and before booting the operating system, you can move these files back to their original locations off the system disk, using the DCL option (8) from the OpenVMS operating system menu..."

We use the following system logical names to relocate files off of the SYS$SYSDEVICEs onto a cluster-common disk:
* LMF$LICENSE
* SYSUAF
* RIGHTSLIST
* NETPROXY
* NET$PROXY
* NETOBJECT
* NETNODE_REMOTE
* QMAN$MASTER
* SYS$SYLOGIN
* VMSMAIL_PROFILE
* VMS$PASSWORD_HISTORY
* VMS$PASSWORD_DICTIONARY
* ACCOUNTNG
* VMS$AUDIT_SERVER
* OPC$LOGFILE_NAME
* VMS$OBJECTS

When I complete the first phase of the rolling upgrade (i.e., upgrade site B's nodes to 8.3), it is not clear whether I need to:

A) Copy Table 4-2 files from updated site B system disk to cluster common disk; i.e., mixed version cluster will use possibly newer/updated(?) files.

B) Delete Table 4-2 files from updated site B system disk and resume using existing files on cluster common disk; i.e., mixed version cluster will use existing ("old") files.

C) Maintain two copies of these files (i.e., site B uses "new" files on updated SYS$SYSDEVICE; site A uses "old" files on cluster common disk) until the site A system disk is updated to 8.3.

D) Some combination of A), B) and/or C).

I have opened a dialog with our HP support team regarding this question, but I'm hoping that some folks here already have experience with a similar scenario (multiple system disks, common system files disk, temporary mixed-version cluster) and can share some feedback.

Thanks,
Dana

[John Gillings · Accepted solution]

Dana,

Closest is B. You need local copies of the files during the upgrade, but when it's complete, revert to the cluster common ones. Obviously don't expect any changes you make to (say) passwords during the upgrade to be reflected in the cluster common data.

This is an area which, sadly, OpenVMS really doesn't do at all well. Every system manager has to roll their own mechanisms for creating and maintaining the cluster common environment. You have to figure stuff like this out for yourself, when it could (and IMHO, SHOULD!) be handled by OpenVMS supplied utilities.

Please make sure you push back on HP to let them know this is an area that needs to be improved!

[Hoff]

AFAIK, there aren't core file structural changes in this range, given these can be used in a mixed-version cluster; the formats are the same.

Do watch for any quota upgrades that can arise in this range.

Haul a copy of these files over to the target system disk, upgrade it, check for core username quota changes and the odd structural change (ANALYZE /RMS /FDL to confirm; but again, I do not believe there are any structural changes here), then (after replicating changes back out to the shared files) resume using the "production" copies. Then start your regression testing. And then start to roll over to the new system disk.

I see the current Cluster SPD oddly neglects to include V8.3 in various of the upgrade matrices. Not sure what's up with that, but it looks like there might be an SPD missing from http://docs.hp.com here, based on what I'm reading in what's posted. (That and there are some product typos in the current SPD. eg: AlphasServer)

Usual reference to the SYLOGICALS.TEMPLATE list here. That's a reasonable list of shared files, and there may well be additions to that list now.

And as HJohn suggests, do push back on HP, both around the (lack of) cluster common tools, and around the upgrade processing. You're doing a normal rolling upgrade here; I'd expect the sequence here to be documented in detail. Either in the manuals, or as a document available from HP support.

[Dana Conroy]

I completed the final phase of our rolling upgrade in December 2009. Thanks for your responses to my question - your feedback was factored into my plan.