RE: Any-one know how to eliminate PLANNED downtime with Oracle RAC?

From: Mark W. Farnham <>
Date: Mon, 24 Nov 2008 19:37:23 -0500
Message-ID: <>

Now Nuno, of course you can upgrade one node while the other nodes keep running. In fact it is quite trivial to upgrade n-1 nodes where you have n nodes in the cluster. It's that last running (and now probably quite overwhelmed node if you really needed RAC in the first place) that gives all the trouble.

Of course I understand that was not really the question you meant to ask...

(and probably you asked it quite precisely at the time if I know you at all.)

-----Original Message-----
From: [] On Behalf Of Nuno Souto
Sent: Monday, November 24, 2008 6:44 PM
Subject: Re: Any-one know how to eliminate PLANNED downtime with Oracle RAC?

Heh! Funny enough: when RAC first came out, I asked Oracle marketing if it would allow me to upgrade one node while keeping the other(s) operational. The reply was: "Yes of course!". Anyone managed to do that yet? No? Nuff said...

You may be spending big for this type of availability. When I was working for the SEM folks, we had this same situation with Google's traffic coming into our servers. We ended up writing extensions to Apache to capture the data we needed into a flat file that kept being rotated, with quite a few RHLinux Apache servers attached to load balancers, all pumping data to these files. The files got collected into a ETL process into our database, which then through an elaborate process provided changes back to the config of the Apache servers as to how to handle the traffic. Whenever a new config was ready, we just sent a signal to the Apache server and the custom code would read it and apply the changes.

Zero downtime front end because of the load balancers, db downtime was acceptable provided we stayed within the window for the feedback loop - around one day. So much so that we didn't even have to use EE, just Standard Edition was enough.

More than one way to skin a cat. I suggest you look at this type of approach rather than "nothing will ever be down". That last one costs a LOT of moolah...

Nuno Souto
in sunny Sydney, Australia

Fahrenfort, Keith (HP Shared Solutions Architect for Oracle) wrote,on my 
timestamp of 25/11/2008 3:25 AM:

> Hi, I'm working with a customer running a critical web site on a 10gR2
> RAC backend DB - they support hundreds of thousands of simultaneous
> connections at the "quietest" time.
> They have expressed a desire for NO downtime during ANY changes to
> Oracle, particularly the application of Oracle patches and Oracle
> upgrades (both minor and major), etc.
> Any thoughts? Who's "been there done that"?
-- --
Received on Mon Nov 24 2008 - 18:37:23 CST

Original text of this message