Re: Availability during software/hardware upgrade.

In article <mWAv9.65258$g9.183303_at_newsfeeds.bigpond.com>, "Howard says...
>
>
>"Stu Charlton" <stuartcharlton_at_hotmail.com> wrote in message
>news:f8851c43.0210290540.b32225d_at_posting.google.com...
>> "Howard J. Rogers" <aldeburgh_at_bigpond.com> wrote in message
>news:<8kit9.59670$g9.169403_at_newsfeeds.bigpond.com>...
>>
>> > Another idea would be to use RAC (Real Application Clusters). You could
>> > have a two node RAC, take one of the nodes offline to patch it/upgrade
>> > it/whatever, and then bring it back into the cluster. Do the same on the
>> > other node, and you achieve truly non-zero downtime for the cluster as a
>> > whole. Again, however, you can't cluster a Windows box with a Unix one,
>so
>>
>> Is this new in 9iR2? I was under the impression from the Oracle 9i
>> RAC course that you can't do a rolling cluster upgrade like this - the
>> whole cluster must be the same version.
>>
>
>I wouldn't want to try an upgrade from 8i OPS to 9i RAC like this, but a
>patch from 9.0.1.1 to 9.0.1.2 would almost certainly be OK. So no, it's not
>new in 9iR2, but there are tolerances.
>
>Regards
>HJR
>

Hate to rain on your parade here, Howard, but as yet there is no guarantee that ANY patch or upgrade can be done in a RAC environment as a rolling upgrade i.e. upgrade one node of the cluster and then the other. The main issue here I believe is that the mind set of the developers needs changing to think about how they can install a patch like this. It's just easier to bring the database down, patch, and then restart it. This is particularly the case when there are changes like datafile header changes, data dictionary changes and so on, where one node of the cluster cannot be out of sync with the other node. I believe (though I could be wrong because I haven't tested this out) that you may be able to do this with a combination of RAC and DataGuard, but even there you could run into problems if the archive log format changed as part of the patch.

Having said all of that, I think this really only causes problems for a handful of people, not the total number that say they need it. The reason I say that is because it's very rare to find a site who has really set up a true HA environment. For example, I've yet to personally come across a site where they run a Test cluster as well as a Production cluster to truly test the impact of a patch in a clustered configuration. THat's not to say they don't exist, but I've had lots of people tell me they need rolling upgrades but who don't have a test cluster.

My $0.42 worth.

HTH. Additions and corrections welcome.

Pete

SELECT standard_disclaimer, witty_remark FROM company_requirements; Received on Wed Oct 30 2002 - 12:19:35 CST