Installing via bioconda installs incompatible version of slurm-jobstep executor

[FelixWodaczek]

Software Versions

snakemake: 9.21.0
conda: miniforge3/25.3.1
snakemake-executor-plugin-slurm: 2.6.1
slurm-wlm: 22.05.8

Full version info after installing using both conda and pip contained in env_list_<pip/conda>.txt.

Describe the bug
When installing the snakemake-executor-plugin-slurm and snakemake-executor-plugin-slurm-jobstep directly using conda, snakemake-executor-plugin-slurm-jobstep is installed in with version 0.4.0.
This leads to all array jobs crashing, as they try to call --slurm-jobstep-array-execs which was not introduced in that version yet.
When explicitly asking for a later version of snakemake-executor-plugin-slurm-jobstep, conda claims that snakemake-executor-plugin-slurm 2.6.1 requires snakemake-executor-plugin-slurm-jobstep 0.4.0.

When using pip to install snakemake-executor-plugin-slurm and snakemake-executor-plugin-slurm-jobstep, snakemake-executor-plugin-slurm-jobstep is correctly installed with version 0.6.0.

Logs
From logs/slurm_logs/60196711_1.log:

__main__.py: error: unrecognized arguments: --slurm-jobstep-array-execs=...

Minimal example

sandbox.tar.gz

Create environments snakemake-slurm-debug-<conda/pip> using

conda create -f environment_<conda/pip>.yml

To run the error producing version run

conda activate snakemake-slurm-debug-conda
snakemake --profile ./slurm_profile.yml --config pkgmgr="conda"

which runs an array job consisting of two jobs executing test.py from the rule_run_test and an additional rule which prints the package information.

Additional context
I've added the slurm log files of both the conda installed version (logs/slurm_logs/60196711_1/2.log) and the pip installed version (logs/slurm_logs/60190718_1/2.log).
env_list_<conda/pip>.txt and env_info_<conda/pip>.txt contain the output of the system_information rule, env_list_<conda/pip>.txt specifically the conda list dump.

[cmeesters]

Thank you for bringing this to my attention. I am a bit behind with all my tasks, but have a release pending and checking for this glitch and adding a correction won't take too long - I see, that I can do it tomorrow.

[cmeesters]

I think this should be fixed in v2.7.0, which is just out. It will take a little while to show up on Bioconda (PyPi is usually faster).

[FelixWodaczek]

Hi Christian,

thanks for looking at this so quickly!
I just tested with v2.7.0, still the same issue when installing from bioconda.

For some reason the recipe generated by the BiocondaBot keeps still has slurm-jobstep v0.4.0 in its dependencies.

Wish I could actually help out more, but to be honest I don't quite know the pipeline for how the recipe is auto-generated on bioconda.

[cmeesters]

Hi Felix,

this is weird. Seems like I need to intervene manually.

I'm afraid, tomorrow and on Monday are public holidays here, and I have some family business. But I will work on it. Meanwhile, you can install the fitting jobstep-version yourself. However, I am currently working on a GPU-reservation bug fix. If this is required too, a fix will probably be released during the next week.

[FelixWodaczek]

No worries, the pip install works problem-free, so it shouldn't be holding anyone back.
Definitely not something to worry about on a long weekend 🙏 from my side at least there is no rush, if others struggle they should at least be able to see this open issue :)