[Pacemaker] Stagger resource migration times

[Florian Haas]

Hi Nick,

On 2011-11-07 18:47, Nick Morrison wrote:

> I'm having trouble understanding pacemaker's timing.  If I type "crm
> node standby host-1", will crm attempt to do everything as quickly as
> possible (shut down all VMs at once)? 

Not quite. On any node, lrmd will only allow a certain number of
operations (start, stop, monitor, etc.) to run in parallel. By default
this number is 4; but it may be modified by passing the
LRMD_MAX_CHILDREN environment variable to lrmd. For Pacemaker 1.1 with
pacemakerd, this setting goes into /etc/sysconfig/pacemaker on RPM based
platforms, and into /etc/default/pacemaker on .deb systems.

While sending a node into standby, the lrmd most likely has to perform
more than LRMD_MAX_CHILDREN actions. Then, it will schedule the first
batch and then line the others up in sequence: as soon as one operation
completes, the next can run.

> How does its resource
> management timing work in this regard?  Is it possible to specify a
> "stagger" or "jitter" time for a deliberate delay between
> stopping/starting VMs on the same host so they don't all compete for
> host resources at the same time?

Is the functionality explained above sufficient for your definition of
"stagger"?

Cheers,
Florian

-- 
Need help with Pacemaker?
http://www.hastexo.com/now