Can systems in HA configs be on two different O/S releases?

[asorrell]

This is a response to a question that came to me in a PM earlier from egsalayon, re:

I've read from your post that you and your team have implemented HA on Linux, is it possible that the OSes of the two DS server have slight difference from one another? Example, 1st DS server OS: RHEL 4.7 AND 2nd DS server OS: RHEL 4.8.

According to IBM, having two systems at different releases is "not recommended". Their documentation expressly recommends "uniform configurations". They want the two systems to be exactly identical, mainly to make it easier to diagnose issues when they arise. I'm not certain that they'd refuse to support you, but they'd definitely be unhappy about the mismatch if you had issues.

With that said, I don't think Information Server cares as long as the version of DataStage on both boxes is identical. However, your HA software package (which monitors system "heartbeats" and does the failover if required) might require the O/S to be identical - you'd need to check with them as well.

However, I don't profess to be an HA expert, just someone that's configured or worked on some HA systems. If anyone else want's to chime in with more info that would be welcome.

Andy

[roy]

Hi,
I wonder why you might want to do this in the first place

[jwiles]

With that said, I don't think Information Server cares as long as the version of DataStage on both boxes is identical. However, your HA software package (which monitors system "heartbeats" and does the failover if required) might require the O/S to be identical - you'd need to check with them as well.
Andy

Andy,

Information Server WILL care!

From a client software perspective, differences in versions (FixPacks, patches, major releases) between the client and the server can be problematic at best and fatal at worst. Imagine a large development effort with tens to hundreds of developers. If a failover occurs and the IS version on the secondary machine is different, you're asking for and will likely receive much trouble.

From a server perspective, there can be major repercussions due to differences in the repository structures, differences in internal logic and so on, even at a FixPack level with a major version. At a minimum, you would somehow need to provide two completely separate repositories, one for each version, and somehow keep them exactly in sync to even have a slim chance at the failover system working at all.

From an engine perspective (your point), maybe not as much as client<>server at first glance. You MIGHT be able to get away with the 8.1 FP2 engine running with an 8.1 FP1 IS server, but I would expect to encounter problems and would not trust the results. I certainly wouldn't recommend this with different major versions (8.0, 8.1, 8.5).

This is just a highlight of potential issues...I'm know there are too many specifics to begin listing here and I doubt there is a comprehensive single list of everything that could be encountered in this type of scenario. Mixing versions is not recommended for good reason....

[asorrell]

James,

I think you misunderstood my reply - the reply was regarding identical UNIX releases, not Information Server releases. I know in the past at two sites I've seen working failover configurations where IIS was identical though the underlying O/S was patched slightly differently.

By the way - I don't disagree with the tenor of your statement. Even if you can get away with having different UNIX releases on the boxes, I do not recommend it. I happen to agree with IBM's stance - the boxes need to be identical from a software perspective. O/S, IIS and all the fixpacks for both!

[jwiles]

Thanks for pointing that out to me, Andy! I somehow completely missed the O/S point of your post. Maybe my answer will be useful for someone anyway

Regarding the different O/S versions and Information Server: yes, it's definitely not recommended for the reasons you point out but it can work well in many cases so long as you're within the same major release and not far apart in point releases (4.7/4.8 in your example). The biggest issue would be shared object/library differences--changes in functionality/results and changes in the expressed function interfaces. Usually that's not an issue in a minor point release, but the slim chance is always there (bug fixes, enhancements, etc.).

Normally you wouldn't expect an organization to want to run two different O/S releases in an HA'd environment, but given economics and corporate budgets, as well as the occasional oddball customer who just does it different, you never know what you'll run into.

Regards,