[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Condor-users] Queued jobs not running on the newly added node, until a new job is submitted

> Hi all,
> I want to build a on-demand cluster, where i can add nodes to the pool
> whenever needed.
> It seems to me that existing jobs in the queue shown in condor_q will
> not be run on the newly added machine (i.e. by condor on <hostname>)
> until a new job is submitted. (i.e. not until a new job is submitted
> does any job will be run on the newly added node). It seems that it is
> because of all jobs are rescheduled once a new job is added to the
> queue.

This is incorrect. The negotiator will use newly added machines on each
negotiation cycle if the machines have had a chance to tell the
collector that they exist. It adjusts the list of available machines
dynamically (which accounts for both new machines joining the pool and
machines dropping out). Depending on your setup there can be a lag of
several minutes between when a machine comes online and when it tells
the collector daemon it exists so the negotiator can use it in the next
negotiation cycle.

> Is there any way to let condor aware of the newly added machine and
> run jobs on it even there is no newly submitted jobs?

This should Just Happen(tm). Albeit with a possible lag as system state
information is collected and made current for the negotiator to use.

You can force a new negotiation cycle by issuing a 'condor_reschedule'
command on any schedd in your system.

You can also shorten the time between negotiation cycles and the time
between startd to collector updates to help new machines come online

- Ian