[Pacemaker] STONITH Deathmatch Explained

Andrew Beekhof andrew at beekhof.net
Tue Aug 11 13:44:33 UTC 2009


On Tue, Aug 11, 2009 at 3:39 PM, Lars Marowsky-Bree<lmb at suse.de> wrote:
> On 2009-08-11T00:41:42, Tim Serong <tim at wirejunkie.com> wrote:
>
> Hi Tim,
>
> it's worth noting that we hunted down an additional new scenario which
> could cause this to trigger.
>
> Some switches seem to take noticably longer to get multicast
> communication up, while unicast or even broadcast are already
> operational. So despite being able to ping, OpenAIS can't see the other
> side and goes into this fencing loop.
>
> This can, of course, be avoided by setting stonith-action=poweroff or
> disabling OpenAIS from being started at boot, but the best fix is to fix
> the switches. ;-)
>
> heartbeat had an "initdead" parameter for the very first initial
> wait time to pronounce the other side dead. Maybe something similar
> would become OpenAIS as well.

does openais care?
it always forms a membership of 1 before even looking for other nodes.

at the pacemaker level there is the dc-deadtime option though, so the
new node will at least wait a while before trying to shoot anyone.




More information about the Pacemaker mailing list