> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:linux-ha-
> [EMAIL PROTECTED] On Behalf Of Andrew Beekhof
> Sent: Tuesday, March 18, 2008 3:25 AM
> To: General Linux-HA mailing list
> Subject: Re: [Linux-HA] trying to understand failback cause - log
> attached
>
> On Tue, Mar 4, 2008 at 4:13 PM, Damon Estep <[EMAIL PROTECTED]>
> wrote:
> > I have a 2 node cluster, heartbeat version 2.1.3, crm=yes,
symmetric
> > cluster=yes, default resource stickiness=50, default resource
> failure
> > stickiness=0
> >
> >
> >
> > The desired behavior is that resources stay where they are unless
> there
> > is an error starting one of them (like v1 auto_failback=off). My
> > understanding is that a positive default resource stickiness value
> > should achieve this (with all other scores being equal).
> >
> >
> >
> > I have attached the ha-log from hb_report in hopes that someone can
> > identify why the actual behavior is this;
> >
> >
> >
> > Prior to 3:00AM the master drbd role and resource group were
happily
> > running on node cn4-inverness-co
> >
> > At 3:00AM a cron event fired that called #shutdown - r now, this
> event
> > was intended to test failover to the other node. I did in fact get
a
> > successful failover.
> >
> > After cn4-inverenss-co came back up and joined the cluster a
> failback
> > from cn3 to cn4 took place, which is not desired.
> >
> >
> >
> > Can anyone provide any clues that might help me understand why this
> > happened and how to prevent it? I can duplicate it consistently on
3
> > different 2 node clusters with the same configuration.
>
> Without seeing your config, I'm guessing that the score in one or more
> of you resource location constraints exceeded the value you set for
> default resource stickiness (ie. 50)
>
[Damon Estep]
That is correct, however I could not overcome it, it appears that the
DRBD Master/Slave resource agent has some internal scoring manipulation
that is difficult to understand, and conflicted with the scoring
manipulation I needed to achieve. I had to switch to a V1 style cluster
and turn off autofailback to get predictable results.
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
