condor_vacate_job

Stop running job(s) and relinquish resources back to the host machine(s) executing the job(s).

Synopsis

condor_vacate_job [-help | -version]

condor_vacate_job [OPTIONS] [cluster… | cluster.proc… | user…]

condor_vacate_job [-debug] [-long] [-totals] [-all] [-constraint expression] [-pool hostname[:portnumber] | -name scheddname | -addr “<a.b.c.d:port>”] [-fast]

Description

Vacate job(s) from the host machine(s) where they are currently running; relinquishing resources back to the host machine(s) and returning the job(s) back to idle to restart resource matchmaking and execution elsewhere. Vacating the job causes the condor accountant to stop charging the user in terms of priority. For any given job, only the owner of the job or one of the queue super users (defined by the QUEUE_SUPER_USERS macro) can vacate the job.

Options

-help
Display usage information.

-version
Display version information.

-long
Display result ClassAd.

-totals
Display success/failure totals.

-fast
Perform a fast vacate and hard kill the jobs.

-pool hostname[:portnumber]
Specify a pool by giving the central manager’s host name and an optional port number.

-name scheddname
Send the command to a machine identified by scheddname.

-addr “<a.b.c.d:port>”
Send the command to a machine located at “<a.b.c.d:port>”.

-constraint expression
Vacate all jobs which match the job ClassAd expression constraint.

-all
Vacate all the jobs in the queue.

cluster
Vacate all jobs in the specified cluster.

cluster.process
Vacate the specific job in the cluster.

user
Vacate jobs belonging to specified user.

General Remarks

If the -name option is specified, the named condor_schedd is targeted for processing. Otherwise, the local condor_schedd is targeted.

Vacated jobs will be sent a soft kill signal (SIGTERM by default or the value of KillSig in the job’s ClassAd) unless -fast is used.

Using condor_vacate_job on jobs which are not currently running has no effect.

Jobs running in the local or scheduler universe can be vacated. Given these types of jobs run directly on the Access Point, they will always restart on the same resources and may begin executing again immediately if there is a low load or number of jobs submitted to the Access Point.

DAGMan can also be vacated and will enter recovery mode upon restarting. However, all jobs submitted and managed by DAGMan will continue running uninterrupted.

Warning

Do not confuse this tool with condor_vacate. condor_vacate is intended for use by the owners/administrators of an Execution Point to evict jobs from their host machine.

Exit Status

0 - Success

1 - Failure has occurred

Examples

To vacate a specific job:

$ condor_vacate_job 23.0

To vacate a job fast:

$ condor_vacate_job -fast 23.0

To vacate all jobs owned by user Mary:

# condor_vacate_job mary

To vacate all vanilla universe jobs owned by Mary:

# condor_vacate_job -constraint 'JobUniverse == 5 && Owner == "mary"'

Availability

Linux, MacOS, Windows