Interconnect switch failure - was: RAC on OCFS2 acceptance testing

From: Boris Dali <boris_dali_at_yahoo.ca>
Date: Mon, 15 Jan 2007 09:20:16 -0500 (EST)
Message-ID: <76649.63143.qm@web32401.mail.mud.yahoo.com>

One of the surprising (at least to me) results of these tests was that while a failure of interconnect NIC (not bonded) on either node of a 2-node RAC doesn't create much trouble, the switch failure triggered a FSFO (fast-start failover) to a target standby.

Just like with a single interconnect NIC failure (on either node) the same voting node goes down with the similar messages...

ospid 13081: network interface with IP address <ip_address> no longer running

Message from syslogd@<node_name>
<node_name> kernel: Kernel panic: ocfs2 is very sorry to be fencing this system by panicing

... yet in contrast to a single NIC failure, RAC (or is it OCFS2?) doesn't recover, failing over to a standby, instead of a master node.

Was that to be expected for a 2-node RAC or is it a problem with our setup (e.g. FSFO-threshold)?

Thanks,
Boris Dali.

Boris Dali <boris_dali_at_yahoo.ca> wrote:

> Thanks, Kevin.
>
> The issue discussed in this SuSE thread is
> documented on Metalink in the note# 394408.1
...

Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com

--
http://www.freelists.org/webpage/oracle-l

Received on Mon Jan 15 2007 - 08:20:16 CST