Hi Marco,
This issue will be treated as a bug. DAGMan should understand when queue is 0/empty as a result of queue for each and decide that no jobs queued at submission time is a successful job run. The fix for this bug will take some time over the course of multiple
tickets just so you know.
Cheers,
Cole Bollig
From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> on behalf of Marco van Zwetselaar <zwets@xxxxxxxxxx>
Sent: Tuesday, February 28, 2023 3:03 AM
To: htcondor-users@xxxxxxxxxxx <htcondor-users@xxxxxxxxxxx>
Subject: Re: [HTCondor-users] Job being queued when script echoes nothing
Thanks John, this is on latest (10.2.0), with direct submit.
I'd argue that a node in a DAG should fail if and only if one or more of its submitted jobs fail. If there are no jobs, then none fail.
The practical application here is rerunning a DAG or batch job where part of the output is already present. A local script listing the remaining work items can then be used in "queue X, Y from ./list-todo.sh |".
More generally: if a job is intended to process a variable number of work items (using "queue ... from"), then it's conceivable that there are 0 items to process. Why treat this as an exception?
Whether DAGman should consider this a success or failure is a separate issue though. What currently happens is that DAGman (contrary to condor_submit, which doesn't produce a job), actually submits the job (once), with empty values for the "queue ..." parameters.
Cheers,
Marco
On 28/02/2023 02:28, John M Knoeller via HTCondor-users wrote:
A DAG is not going to know how to deal with a submit file that does not submit any jobs – should it treat that as a failure? A success?
You do not say what version of HTCondor, or of DAG man. Is this a modern versions that does direct submit? Or an older version that is using condor_submit to submit the jobs?
-tj
Hi John,
This old issue bit me again, so I double checked. What's happening is that the bug occurs only when the job is in a dag.
With the following:
echo-nothing.sh:
#!/bin/sh
exit 0
empty.job:
executable = /usr/bin/echo
arguments = "[$(ARG)]"
output = job.out
queue ARG from ./echo-nothing.sh |
empty.dag:
JOB empty empty.job
Submitting empty.job correctly gives the warning and queues no jobs, whereas "condor_submit_dag empty.dag" surprisingly queues and executes the job.
Cheers
Marco
On 22/09/2022 18:09, John M Knoeller via HTCondor-users wrote:
I think your script must be outputting something because when I try with a script that prints nothing I see this
Submitting job(s)
WARNING: "fromcmd.sub" has only empty "queue" commands -- no jobs queued
Or perhaps there is some complicating factor in your submit file. Could you send me a submit file and script that reproduces the problem?
-tj
Dear all,
I regularly use this construct in my job files:
queue ID, FN from ./list-todo.sh |
e.g. to skip jobs that have already been done.
However when all jobs have been done, and ./list-todo.sh echoes nothing (not even a newline), Condor still queues a job. The ID and FN are set to '', so the job usually fails, whereas IMO it shouldn't have been queued in the first place.
An obvious workaround is to add a wrapper script, but that isn't very clean: the "non-job" still gets scheduled and executed, consuming resources it won't use. Worse, if $(FN) appears in transfer_input_files, then this won't work.
I would think this is a bug, but maybe there's a knob somewhere that changes this behaviour?
Thanks,
Marco
--
Kilimanjaro Clinical Research Institute
P.O. Box 2236 | Moshi, Kilimanjaro | Tanzania
_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users
The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/
_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users
The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/
|