[Pacemaker] migrate_to op parameters ignored?

[Alain St-Denis]

Le 14 octobre 2009 10:20:12, Andrew Beekhof a écrit :
> On Wed, Oct 14, 2009 at 4:18 PM, Andrew Beekhof <andrew at beekhof.net> wrote:
> > try migrate-to :-)
> > we tried to standardize on dashes since 1.0.
> 
> doh, almost right.  right reason, wrong field: you want on-fail not on_fail

on_fail and on-fail both seem to work. I just realized I also need to set on-
fail for the migrate_from op... All is fine now.. I wonder what is the 
usefulness of exposing both migrate ops if they can't be used independently. 
Why not a single migrate op (even if it translates to migrate_to and from 
under the covers)?

I'm still puzzled by the timeout value being overridden by the one given to 
the stop op. Is this the implemented behaviour? If it is, could you please 
explain the rationale behind it? If not, should I open a bug?

Regards.

> 
> > On Wed, Oct 14, 2009 at 2:04 PM, Alain St-Denis <alain.st-denis at ec.gc.ca> 
wrote:
> >> Hi,
> >>
> >> I'm trying to set "on_fail" to "block" for the migrate_to op but it
> >> seems it is ignored. My reasoning is this: if my server hosts 15 Xen
> >> VMs, I don't want one failed migration (for whatever reason) to take
> >> everything with it. Currently, a failed migration causes a stonith and
> >> hence causes an unnecessary reboot of all other VMs on that host. Am I
> >> missing something?
> >>
> >> Also, It seems the timeout for migrate_to is bound to the value set for
> >> the stop operation. Is that right? Why?
> >>
> >> This is with pacemaker-heartbeat 1.0.4 on Debian Etch.
> >>
> >> Thanks in advance.
> >>
> >> --


-- 
Alain St-Denis
Supercomputing, Systems and Storage / Superinformatique, systèmes et stockage,
High Performance Computing Support / Soutien aux calculs en haute performance
Chief Information Officer Branch / Direction Générale du dirigeant principal de 
l'information
Environment Canada / Environnement Canada
Tel: +1 514 421 4697