Re: An outage-free upgrade?

Hi!

Ah, my favourite topic ;)

I've done it the same way with triggers, on a 600GB database, when migrating between platforms.
Another way would be to use snapshot & snapshot logs, or if you don't want to put additional load to you environment, use logminer (which could be somewhat problematic in 8i, because limitedness of 8i logmnr).

Have you checked what is the main performance bottleneck when doing the upgrade?
Usually with these kinds of serial scripts which run a lot of small DDL commands, log file sync will be one of the most seen wait (along with log file parallel write). I've got over it by settinng _wait_for_sync to false. But if your database crashes during this parameter set to false, it is probably corrupt. This is an unsupported parameter, so test it first & be aware of possible consequences.

Tanel.

"Vsevolod Afanassiev" <vafanassiev_at_aapt.com.au> wrote in message news:4f7d504c.0308060010.6505f49e_at_posting.google.com...
> Is it possible to upgrade Oracle database without having
> a long outage? For example, I need to upgrade 8.16 -> 8.1.7.
> I can install the software in advance (it goes into another ORACLE_HOME
> anyway), but running upgrade scripts (u0801060.sql, jvmu816.sql,
catxsu.sql)
> takes more than 1 hour. The database is not
> available during this period.
>
> Is there any way to reduce this period to two short
> outages?
>
> For example, I can set up standby database and use "switchover"
> functionality, but this means that at some point in time I'll need
> to apply redo logs generated for 8.1.6 to 8.1.7 database - most likely
> it won't work...
>
> Even with Parallel Server/RAC all nodes have to run the same version...
>
> So the only solution I could come up is to create triggers
> on ALL application tables that will capture all update/insert/delete
> activity, failover to standby, upgrade primary, and then
> use information captured by the triggers to "replay"
> the changes in the primary.
>
> Thanks,
> Sev
Received on Wed Aug 06 2003 - 07:36:24 CDT