Re: [HTCondor-users] What happens of the USER_JOB_WRAPPER doesn't exec?

[Jaime Frey <jfrey@xxxxxxxxxxx>]

The âmustâ in the documentation is more of a âreally shouldâ. If your wrapper script alters how HTCondor interacts with the job, it can confuse HTCondor and the user. The primary concern here is how signals are delivered to the job and how the jobâs exit status is determined by HTCondor. If youâre careful, your script can handle these properly.

Have you considered have your wrapper script fork()/exec() your monitor/management code and then exec() the user job?

 - Jaime

> On Jun 13, 2025, at 10:49âAM, KÃhn, Max (SCC) <max.fischer@xxxxxxx> wrote:
> 
> Hi all,
> 
> The docs for the USER_JOB_WRAPPER [0] currently state the wrapper cannot persist while the job runs:
> 
>> This wrapper script must ultimately replace its image with the user job; thus, it must exec() the user job, not fork() it.
> 
> However, we would really, really like to use one of forkâs brethren for managing and monitoring stuff while the job runs.
> 
> Is this line from the docs still authoritative? How bad would it be if we donât follow it? Has someone (*gasp*) done this already and still lives to tell the tale?
> 
> Cheers,
> Max
> 
> [0]
> https://htcondor.readthedocs.io/en/latest/admin-manual/configuration-macros.html#USER_JOB_WRAPPER
>