Patching SerivceGuard

[Alex Georgiev]

This should be a simple question for the tried and true ServiceGuard admins.

I'm about to patch an ServiceGuard cluster (A.11.15). This is the most basics SG setup you can imagine - two nodes, one running a production package, the other running a test package of the same app. I'll be installing patches:

PHSS_32245
PHSS_32727

The initial plan was to install these on the non-production (aka standby) node, then let it run for 2 week until I can get the downtime. Then in two weeks, flip the cluster and patch the 2nd node which has been running a production environment until then.

So, the 2 weeks during which these nodes will be at different patch levels... good idea or bad idea?

Looking for opinions, experience, and wizdom.

Thanks much!

[melvyn burnard]

Well there is no real reason why you need do it this way, it should be fine to patch both nodes at the same time, bearing in mind you need to halt the node c;uster activities at the time of install.
However, if you do feel you need to do it this way, that should be fine, just remembering you now have one node with bug fixes that the other does not.

[Stephen Doud]

The cluster will operate, but it is not advisable to attempt to change the cluster or package configuration when the nodes operate on dissimilar patch levels. Errors can be produced from cluster configuration commands in such a state, so expect it.

[Chan 007]

Alex,

I agree with Steve's way. Thats how you do that. Patch level should be same for all nodes.

I know that you are concerned if things gone wring what to do?. If you are so scared about your problme, then keen your root mirror disk apart on the secondary node for a day or 2.

Chan

[Geoff Wild]

I wouldn't run at all for more then 2 hours at different revisions - think of a cluster as a single entity. All nodes should be identical at all times.

If you have issues - just back out the patches.

If I were you, I would seriously look at upgrading to the latest version of ServiceGuard (11.17).

Rgds...Geoff

[John Bigg]

I have to disagree with Stephen, Chan and Geoff. There is a difference between running different versions for an extended period of time during a rolling upgrade, e.g. 11.16 on 1 node and 11.15 on another, and running at different patch levels.

Running on different patch levels should be fine and is not all that unusual. It is expected that you cannot patch all nodes at the same time and if you look back at the history of patches you will see a couple that make changes which only take effect when the patches are installed on all nodes, for example PHSS_26056. It is expected that nodes may be at different patch levels.

You should be able to run with different patch levels and I would not expect this to cause trouble, even if you needed to make configuration changes with cmapplyconf. I would not recommend this for extended periods but I think what you are suggesting is fine.

Even with rolling upgrades where you have different releases (and configuration operations are explicitly barred) the manuals only state, "It is highly recommended that you upgrade all cluster nodes as quickly as
possible to the new release level." and timescales are not quoted. It is not that unusual for nodes to be upgraded days apart.