Re: RAC: typical time to reboot a node and have it rejoin the cluster?

From: John Hurley <johnbhurley_at_sbcglobal.net>
Date: Fri, 19 Jun 2009 15:53:06 -0700 (PDT)
Message-ID: <3ed9090b-c0d3-49c7-86ce-be51aaee6ceb_at_r13g2000vbr.googlegroups.com>



On Jun 19, 6:13 pm, m..._at_pixar.com wrote:

Mark who says he is from pixar noted ...

> What's a reasonable time for a rebooted RAC node to join a cluster?
> If you have a RAC, how long does it usually take for you?
> Is 5 minutes a "reasonable" number?  
>
> I realize this is a very open ended question with a very broad
> range of possible answers, I'm just looking to get an idea
> of what is plausible in order to set my expections accordingly.
>
> Many TIA!
> Mark

Are you counting the time to "reboot" the server in your answer?

It can take several minutes or more for many modern servers to get thru various BIOS startup things. Typically medium scale and up servers do memory checking and internal diagnostics first. Often there is a RAID controller startup phase ... initialization of HBA's ( Hot Bus Adapters ) ... etc ... several minutes typically before linux even gets a chance to start working on it's initialization.

Then you get things with oracle involved eventually as clusterware starts up ... etc.

It can easily take in the range of 5 to 10 minutes ... sometimes longer before a node is back in the cluster.

Too often 2 node RAC clusters experience symptoms where 1 node comes back in and the surviving node then reboots ... some people advocate clusters of 4 or more nodes.

It really seems surprising that someone from Pixar is asking questions like this at an internet newsgroup. Don't you have oracle consultants all over you trying to do free work for you as a way in? I would think Pixar would want to bring in experienced RAC people in house instead of experimenting around home growing stuff.

But anyways ... you really need to test it yourself on the hardware you will be running on to have any answer that is very close to the truth ... it will vary tremendously on different servers ... different storage devices ... different RAC topologies. Received on Fri Jun 19 2009 - 17:53:06 CDT

Original text of this message