[Pacemaker] Two stonith devices

[Dejan Muhamedagic]

Hi,

On Tue, Jun 07, 2011 at 03:15:46PM +0300, Vladislav Bogdanov wrote:
> Replying to myself after studying TODO.
> 
> Could you guys (Dejan, Andrew) please describe possible ways for fixing
> this, may be I'll be able to do it to some extent. I still need to study
> code which is a kinda black magic where it goes to stonith.
> May I ask for some pointers to start with?
> And, is design of stonith subsystem documented somewhere? Can you
> shortly describe it if it is not?

This should work. IIRC, there was something about priorities not
working yet in stonith-ng. For the implementation, you'll need to
ask Andrew, but it may take a while before he gets back to you.
Please see below.

> Best,
> Vladislav
> 
> 
> 04.06.2011 12:08, Vladislav Bogdanov wrote:
> > Hi all,
> > 
> > silly question:
> > 
> > does anybody have working configuration with two stonith paths to
> > cluster node?
> > 
> > Background is:
> > I have two stonith resources configured for each node in two-node cluster:
> > primitive stonith-s01-0 stonith:external/ipmi \
> >         params hostname="s01-0" ipaddr="10.5.4.250" userid="xxx"
> > passwd="xxx" interface="lanplus" pcmk_host_check="static-list"
> > pcmk_host_list="s01-0" priority="1" \
> >         meta priority="2" \
> >         op monitor interval="60"
> > primitive stonith-s01-0-s stonith:rcd_serial \
> >         params hostlist="s01-0" ttydev="/dev/ttyS0" msduration="320"
> > dtr_rts="rts" \
> >         meta priority="5"
> > primitive stonith-s01-1 stonith:external/ipmi \
> >         params hostname="s01-1" ipaddr="10.5.4.252" userid="xxx"
> > passwd="xxx" interface="lanplus" pcmk_host_check="static-list"
> > pcmk_host_list="s01-1" priority="1" \

Why the two pcmk_* parameters? They necessary? Looks silly.

> >         meta priority="2" \

This must be a parameter, meta is superfluous.

> >         op monitor interval="60"
> > primitive stonith-s01-1-s stonith:rcd_serial \
> >         params hostlist="s01-1" ttydev="/dev/ttyS0" msduration="320"
> > dtr_rts="rts" \
> >         meta priority="5"
> > 
> > I tried to simulate node failure (killall -9 corosync) when ipmi devices
> > are inaccessible. First try was successful, s01-0 was fenced by
> > rcd_serial. s01-1 was not fenced when I killed corosync there. Logs do
> > not show any attempt to use stonith-s01-1-s.
> > 
> > Am I missing something? (Yes, I verified that rcd_serial works, I fenced
> > s01-1 with it manually.)
> > 
> > pacemaker-1.1.5 with several patches cherry-picked from 1.1 and devel
> > branches.
> > corosync-1.3.1

Best to file a bugzilla, it seems to me.

It would be good to know if there's anybody using multiple
stonith devices with pacemaker 1.1.x.

Thanks,

Dejan

> > Best,
> > Vladislav
> > 
> > _______________________________________________
> > Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> > http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> > 
> > Project Home: http://www.clusterlabs.org
> > Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> > Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
> 
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker