Re: Another newbee RAC question

I can't speak to 12c or some of the new features there (Transaction Guard, Application Continuity, etc), but through 11gR2, existing connections will not relocate automatically. Services can be relocated using 'srvctl', but even if you do that, only future connections will be affected. Existing connections won't move, unless you specify the '-f' flag to 'srvctl relocate service'. If you do specify the '-f' flag, it just disconnects all the connections for the affected service. The application would have to be coded to catch the disconnection error and reconnect. All this is probably for the best, as failover can be considered disruptive, and there's no need to be unnecessarily disruptive. There's new stuff in 12c that may make this better, but I'm not up to speed on all of that, yet.

If you're concerned because, for example, you have a two-node RAC, and you lose a node, and you're worried about the remaining node having adequate capacity to support all the users, then, I'd argue that you don't have N+1 redundancy in terms of capacity, and that your HA configuration is not really valid. If you're not running RAC for true HA, then why not just get one big box, and run single-instance, avoiding RAC and the attendant costs and overheads altogether?

Hope that helps,

-Mark

PS If you want to automate relocating your services, and you know how to write a shell script to do what you want, you could hook it to clusterware, via a user event callout, see $GRID_HOME/racg/README for details. (Available in 11gR2, not sure about earlier versions.)

From: Chris King <ckaj111_at_yahoo.ca<mailto:ckaj111_at_yahoo.ca>> Reply-To: Chris King <ckaj111_at_yahoo.ca<mailto:ckaj111_at_yahoo.ca>> Date: Friday, May 16, 2014 at 12:01 PM
To: "oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org>" <oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org>> Subject: Another newbee RAC question

First, thanks everyone for all your assistance with my many newbee RAC questions. You've been a huge help!

I just did my first failover test, and everything looked fine. When I brought the failed rac node up again, the client connections were still all connected to the node that did NOT fail. From what I read in Oracle's FAQ, "all the types of load balancing available currently (9i-10g) occur at connect time. I'm running 11gR2.. Is this still the case? Or is there a way to re-balance client connections in 11gR2 once a node rejoins the cluster.

Thank you!

--

http://www.freelists.org/webpage/oracle-l Received on Fri May 16 2014 - 20:16:13 CEST