[Linux-HA] About time-out of STONITH.

[Andrew Beekhof]

On Tue, Jun 17, 2008 at 19:04, Dejan Muhamedagic <dejanmm at fastmail.fm> wrote:
> On Tue, Jun 17, 2008 at 03:45:10PM +0200, Andrew Beekhof wrote:
>> On Tue, Jun 17, 2008 at 15:07, Dejan Muhamedagic <dejanmm at fastmail.fm> wrote:
>> > On Tue, Jun 17, 2008 at 08:49:52AM +0200, Andrew Beekhof wrote:
>> >> On Tue, Jun 17, 2008 at 08:14, HIDEO YAMAUCHI
>> >> <renayama19661014 at ybb.ne.jp> wrote:
>> >> > Hi,
>> >> >
>> >> >> Which operation did you cause to fail?  The monitor or the fencing operation?
>> >> > When I caused fencing, I generated time-out in RA of STONITH.
>> >> >
>> >> > Possibly is this phenomenon improved in Pacemaker?
>> >>
>> >> By design the crm has no idea which node or plugin is used (or were
>> >> tried and failed).
>> >> So it would be impossible to for it recover the RA on its own.
>> >>
>> >> It _may_ make sense for the stonithd to unload any plugin that reports
>> >> a failed stonith action (and thus causing the monitor to eventually
>> >> fail) but I'll leave that for Dejan to think about :-)
>> >
>> > stonithd doesn't unload any plugins depending on failed actions.
>>
>> right - it was a suggestion for a possible enhancement.
>> no idea if its a good idea though.
>
> It would save some memory, not much though, but it'd also
> increase the code's complexity. The latter probably by far
> outweighs the former :)

The other benefit is that it would force the monitor to fail - but
perhaps that just means the monitor isn't good enough...