Re: Availability during software/hardware upgrade.

Another flurry of useful replies, I see!

Yes, what you ask for is possible, provided you have version 9i, and provided you implement something called Data Guard (ie, Standby Database on steroids).

You have a requirement to switchover to Database 2, and then back again to Database 1: that's not possible with any version of Standby Database before 9i. You *could* "failover" to Database 2 in previous versions, but that caused Database 2 to receive a resetlogs, meaning it was in a new incarnation compared to Database 1, and meaning that you couldn't failover back to Database 1, until Database 1 had been completely blown away, re-cloned from Database 2, and the standby re-initialised.

Havng said that, any and all versions of Standby, including Data Guard, require both Databases to be running on the same operating system, so it's not possible to use this mechanism to do a rolling upgrade from, say, Windows to Unix. And the doco. states that each O/S must be at the same patch release, so point upgrades of the same O/S are theoretically not possible either.

The costs of Data Guard are considerable. Not only must you be running the Enterprise Edition, but to satisfy your requirement to lose no records, you'll have to be running in one of the three higher modes of protection which make LGWR do the transmission of redo from one Database to the other. And that has significant performance implications (nothing that some expensive kit won't fix!!).

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 O/S changes are out of the question. Point release upgrades of the same O/S shouldn't necessarily be a problem, however.

Also, I note you require no 'broken' transactions. Data Guard requires that you ask your users to get off the database nicely, so in theory you shouldn't break anything there. But RAC cannot failover DML (it can transparently failover selects, but not inserts, updates or deletes). So anyone connected to a RAC node that you cause to leave the cluster would lose any DML they were in the middle of.

Anyway: that's two things for you to investigate: Data Guard and RAC. Both require the Enterprise Edition; both are expensive; both require good hardware; both are quite tricky maintenance prospects for the DBA.

HTH
Regards
HJR "Andre Sluiter" <A.Sluiter_at_gesloten.nl> wrote in message news:3db50850.3320984_at_news.nl.uu.net...
> Is it possible to change over from
> hardware, OS version, Oracle version
> without going off line ?
>
> I am thinking in the line of two (or more)
> servers where server 1 is online, server 2 gets
> upgraded, then a switch over from 1 to 2
> server 1 get's upgraded and then a change back
> to server 1.
>
> The switchover should be done without dataloss
> or broken transactions.
>
> Is this possible ?
> What is required for this ?
> (If you can put a price on this please do,
> or point me to a website or more info, thanks).
>
> Just checking if we should be looking into this.
> (I have always claimed that if feasable this is
> going to be to expensive for us, but have no
> solid arguments).
>
>
> Andre Sluiter
Received on Tue Oct 22 2002 - 15:12:04 CDT