Hi all,

A couple of weeks ago I added partial support for -M/--clusters to
slurm-drmaa, however, this did not include support for providing a cluster
list on submission (i.e. `sbatch --clusters=cluster1,cluster2`). In its
current implementation, only 0 or 1 cluster is supported.

I've recently been working on adding the missing functionality. I started
with the public slurm_job_will_run() function, but it turns out this does
not provide enough, as sbatch's mult_cluster.c reveals: "We don't use the
api here because it does things we aren't needing like printing out
information and not returning times."

I have it working in development by essentially copying the code from
mult_cluster.c into slurm-drmaa with a few modifications, but it also
requires a copy of the Slurm source at compile time due to a lot of that
code using types, constants, functions, etc., which are defined in private
headers. And gethostname_short() is not exposed in libslurm.

So, could a public API function be added that performs slurm_job_will_run()
but that also provides access to start times, as _job_will_run() does in
mult_cluster.c? I'd be happy to submit a PR for this, especially if there's
a chance of it making it into 14.11.

Thanks,
--nate

Reply via email to