[Linux-HA] Don't start all groups at once?

[Dejan Muhamedagic]

Hi,

On Mon, Oct 01, 2007 at 06:04:41PM -0700, Kelly Byrd wrote:
> As I've mentioned before, I've got 22 groups, each of which is tied to one
> of  22 master/slave drbd resources. During normal operation, one node runs
> 11 and the other node runs the other 11.
> 
> During a failure, 11 drbds their 11 groups will start on the remaining
> node. This will be quite a load on the surviving machine and it's
> currently running resources. Each group is a virtual machine trying to
> allocate a few hundred megs of RAM and boot an OS. I'd like to space it
> out a bit. Is this possible? I don't want to specify an strict ordering. I
> don't actually care about the order and I don't want VM-14 to not come up
> because VM-13 failed for some reason. Ideally, I could configure things so
> heartbeat will start a few at a time, then wait a bit, then start a few
> more. If that's not possible, then starting them all sequentially would be
> fine, and I'll just put a delay in the group. I'm under the impression
> that heartbeat will try and start all groups at the same time. Is this
> true?

Yes and no. One part, the CRM will want to have all started
at once. However, the other one, the LRM doesn't run more
than four resources at a time. If there are four currently
running and some more runnable the latter are postponed for one
second and then the queue checked again. This is, however, an
undocumented feature, so it may change in the future. At any
rate, we do try not to overwhelm the host. Note that the LRM has
no notion of how "heavy" processes are. It would be interesting
to see the results with this multitude of VMs as resources.

To be on the safe side, as Andrew suggested, you should use
various start-delay periods.

Thanks,

Dejan

> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems