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

Re: [Condor-users] Setting the master and startd for one job only

Marian ZUREK wrote:

I would like to setup the condor worker node to run precisely one job and exit. For this to happen I got the receipt to start the condor_master in the foreground (-f option)
and also define the DAEMON_SHUTDOWN expressions as specified below

STARTD.DAEMON_SHUTDOWN = (Claimed && Idle && TotalTimeClaimedBusy > 0)
MASTER.DAEMON_SHUTDOWN = startd_exit_code == 99

I have tested the above setup using Condor v6.9.4 however I did not manage to make it operational the desired way. I am wondering whether any of you had a chance to exercise such a scenario and is willing to share his/her experience.

So you want the startd and master to exit after the startd runs just one job to completion?

Hmmm.... well, I haven't tested this, but off the top of my head this is what I'd try appending to the end of condor_config:

# Tell the master to wait a real long time before restarting the startd,
# since when the startd exits we don't want to restart it, we want to go
# away.

# Set CLAIM_WORKLIFE to one second.  This tells the startd to go back
# to Unclaimed/Idle state after any job that runs for more than one
# second completes.

# Modify Start expression so we will not accept a second job after
# the first one completes and the startd goes back to Idle.
START = (TotalTimeClaimedBusy =?= UNDEFINED || \\
         TotalTimeClaimedBusy =?= 0 ) \\
         &&  ( $(START) )

# Shutdown startd if we are currently idle and have ever been busy
# in the past.
STARTD.DAEMON_SHUTDOWN = (Activity=="Idle") && \
   ((TotalTimeClaimedBusy =!= UNDEFINED) && (TotalTimeClaimedBusy > 0))

# Shutdown the master if the startd exits.  Note that if
# STARTD_StartTime is zero, that the startd is not currently running.

Thank you in advance for your replies,
Marian ZUREK

Hope this helps Marian. If it does, you can buy me a beer next time you are in Madison. :).

Todd Tannenbaum                       University of Wisconsin-Madison
Condor Project Research               Department of Computer Sciences