Please advice!

[dba_dada]

Linux Redhat 8 and Oracle 8i or 9i.

I have to build a failover server in read/write environment automatically in
the event of a system failure. This application being business critical,
we would like to have a complete and synchronous on another database located
at another site (the 2 sites are linked via t1).

There would be only one master, and we would have to
switch automatically to the alternate database in case of problem.

I was hoping that you could share
your experiences and approaches to this problem.

Thanks!

[jovery]

If you need to have read/write access to both databases then your only option is replication. If you are simply looking for a hot standby solution then you can use Oracle Dataguard.

regards

[Tarry]

Well, as jim stated you're(if i understand correctly) looking for a standby database/or in 9i called as a dataguard in managed recovery mode(meaning archived logfile's being shipped automatically) there are certain parameter's you'll have to setup to decide the urgency of the shipping and/or validation of block corruptions(so that before a block corruption) reaches the standby database, you get to know.
but still better read the doc's, it's very interesting.

[Sureshy]

Not sure about standby config in 9i but assume it is near enough same concept as 8i.

Your standby databse is only as up to date as the last archive log file it received. If your primary database crashes before you are able to archive its current redo log file then your standby database will be behind by that amount.

I had to consider this when designing a failover system for one of our 24/7 servers, I have decided to use multimaster repication as this will provide a more up to date failover system than standby.

Please don't ask me about replication configuration as it is a massive subject which I have only just started looking into.

[jmodic]

Originally posted by Sureshy
Not sure about standby config in 9i but assume it is near enough same concept as 8i.

Your standby databse is only as up to date as the last archive log file it received. If your primary database crashes before you are able to archive its current redo log file then your standby database will be behind by that amount.
ubject which I have only just started looking into.

Not any more with 9i. In 9i you can configure Dataguard (aka standby database) so that you don't loose any transaction at all. Not a single one!

[Calvin_Qiu]

Though I think that Dataguard would be the most suitable solution in this case, only one point all you guys ignored, how the Dataguard could do a failover automatically by itself?

You probably need to write some code to monitor the status of the Dataguard, then do a failover when you need it, you also might wanna configure the clients to TAF mode.

[Sureshy]

Automatic failover can be configured through sqlnet/tnsnames can't it ?

[marist89]

Connect-time failover, yes. Transparent Application failover, no.

[ramses71]

A couple of things here.

If you want to use Dataguard in 9i (9.0.2 specifically) you have the recourse to a logical standby database which can be updated as per your specific requirements (default is on the log switch - they have a feature called DSO which replicates the data through db links on the fly per transaction though I have not played with that yet). Also, with the logical it can be open for users to do their reporting on and they can update tables that are NOT part of the target tables for Data guard.

Failover is still manual though there is something that you can consider here - RAC with Cluster Guard. You get the failover (automatic) with standby databases if you need. This is a nasty setup though. Data Guard is relatively easy to setup but RAC is a pain but you do get your high availability with boths options here. Just a thought though.

[dba_dada]

Thank you all.