Open meliamne opened 3 months ago
Actually, the idea is: provide the most-common flags, wrap everything else in slurm_extra
(documentation is forthcoming).
However, I will discuss adding cluster
and gres
now.
Thank you very much for looking into this.
I think the current approach generally makes a lot of sense. The main issue in our HPC's case is that, as far as i can tell, the --cluster
option from slurm_extra
is only applied when submitting jobs. However, it is not applied when checking the status of a job with sacct
or canceling a job with scancel
. In our environment running these commands without the --cluster
option specified returns nothing. Thus after starting the first round of jobs, snakemake gets stuck trying to check the status of the jobs.
Please check, whether PR #56 fixes your issue: I do not have access to a multicluster setup. That might change in the future, but for now, I cannot test this feature.
And you are right: Checking the job status and cancelling jobs, has to be done be the executor plugin. If the information is only available at submit time, this would not be sufficient.
Hi,
Our hpc runs 2 slurm clusters, this requires us to provide the -M or --cluster option to most slurm related commands like sbatch, squeue and scancel. This option is unsupported in snakemake 8+ as far as i can tell. In older snakemake versions support existed through profiles.
I can get submission to work by specifying --cluster in slurm_extra. However, snakemake is unable to detect when a job finishes and start the next set of rules.
slurm: 23.02.7 snakemake: 8.9.0 snakemake-executor-plugin-slurm: 0.4.2