On Tue, Jan 10, 2012 at 10:20:02AM +0100, Andreas Kurz wrote:
> Hello,
> 
> On 01/06/2012 06:14 PM, erkan yanar wrote:
> > 
> > Moin,
> > 
> > Im having the issue, that promoting a master can run into the promote 
> > timeout.
> > After that, the resource is stopped and started as a slave.
> > 
> > In my example it is a mysql resource, where promoting is going to  wait for 
> > any replication lag to be
> > applied. This could last a very long time.
> 
> If the resource is not ready be promoted it should have no promotion
Agree, thats why Im trying to trick with cibadmin.
> score ... is this the mysql RA coming with the resource-agents package
> or a home-grown RA?

The RA is homegrown. As a fact Ive got to wait to see if the RA from Yves
Trudeau is going to get into the official RA. So trying to commit now my own
changes, could be to early. On the other hand I still don't get the 
notify issue. At least for my workload. And of course learning writing that RA
I feel confident to push relevant code.


> 
> > 
> > There are some thoughts on that issue:
> > 1. Dynamically increase the timeout with cibadmin. I havent tested that 
> > yet. Would this work?
> > 2. op-fail=ignore
> > With ignore, the resource is not restarted. But I don't like that approach.
> > 
> > Is there an intelligent approach to dynamically change the timeout while 
> > promoting?
> > Or is there a better approach anyway?
> 
> Even if it would be ok to promote such an instance, why not increasing
> the promote timeout to a "fixed" safe value?

What is a safe value?
The reason I would like to change the timeout is only if the slave lags. (You
can configure the behaviour what to do if the slave lags.) For every other
reason promotion should fail. A fixed (to high) value would count against all
other issues to.


Regards
Erkan


-- 
über den grenzen muß die freiheit wohl wolkenlos sein
_______________________________________________
Linux-HA mailing list
Linux-HA@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to