Search code examples
bashshellslurmsnakemake

Conflicts between Snakemake and GROMACS?

I tried to simplify my issue as much as possible and still getting the error.

The whole idea is that I want to execute (inside a much more complex workflow) the command: gmx mdrun -nt 12 -deffnm emin -cpi on a cluster. For that I have a conda environment with GROMACS and Snakemake.

On a traditional way I have the jobscript (traditional_job.sh) with:

#!/bin/bash
#SBATCH --partition=uds-hub 
#SBATCH --nodes=1
#SBATCH --cpus-per-task=12 
#SBATCH --mem=5000
#SBATCH --time=15:00
#SBATCH --job-name=reproduce_error
#SBATCH --output=reproduce_error.o
#SBATCH --error=reproduce_error.e

gmx mdrun -nt 12 -deffnm emin -cpi

And everything works as expected after sbatch traditional_job.sh. However, if I try to use Snakemake instead, the problems start.

My Snakefile is:

rule gmx:
    input:
        tpr = "emin.tpr"
    output:
        out = 'emin.gro'
    shell:
        '''
           gmx mdrun -nt 12 -deffnm emin -cpi
        '''

And my job.sh:

#!/bin/bash
snakemake \
    --jobs 10000 \
    --verbose \
    --debug-dag \
    --latency-wait 50 \
    --cluster-cancel scancel \
    --rerun-incomplete \
    --keep-going \
    --cluster '
        sbatch \
        --partition=uds-hub \
        --nodes=1 \
        --cpus-per-task=12 \
        --mem=5000 \
        --time=15:00 \
        --job-name=reproduce_error \
        --output=reproduce_error.o \
        --error=reproduce_error.e '

After ./job.sh, the GROMACS's error (written out on reproduce_error.e) is:

Program:     gmx mdrun, version 2022.2-conda_forge
Source file: src/gromacs/taskassignment/resourcedivision.cpp (line 220)

Fatal error:
When using GPUs, setting the number of OpenMP threads without specifying the
number of ranks can lead to conflicting demands. Please specify the number of
thread-MPI ranks as well (option -ntmpi).

For more information and tips for troubleshooting, please check the GROMACS
website at http://www.gromacs.org/Documentation/Errors

I observed that in the output of snakemake is written a different shell (#!/bin/sh). But honestly, I do not know if that could be the problem, neither how to solve it if that is the case:

Jobscript:
#!/bin/sh
# properties = {"type": "single", "rule": "gmx", "local": false, "input": ["emin.tpr"], "output": ["emin.gro"], "wildcards": {}, "params": {}, "log": [], "threads": 1, "resources": {"mem_mb": 1000, "mem_mib": 954, "disk_mb": 1000, "disk_mib": 954, "tmpdir": "<TBD>"}, "jobid": 0, "cluster": {}}
cd '/home/uds_alma015/GIT/BindFlow/ideas/reproduce error' && /home/uds_alma015/.conda/envs/abfe/bin/python3.9 -m snakemake --snakefile '/home/uds_alma015/GIT/BindFlow/ideas/reproduce error/Snakefile' --target-jobs 'gmx:' --allowed-rules 'gmx' --cores 'all' --attempt 1 --force-use-threads  --resources 'mem_mb=1000' 'mem_mib=954' 'disk_mb=1000' 'disk_mib=954' --wait-for-files '/home/uds_alma015/GIT/BindFlow/ideas/reproduce error/.snakemake/tmp.lfq5jq4u' 'emin.tpr' --force --keep-target-files --keep-remote --max-inventory-time 0 --nocolor --notemp --no-hooks --nolock --ignore-incomplete --rerun-triggers 'software-env' 'params' 'input' 'mtime' 'code' --skip-script-cleanup  --conda-frontend 'mamba' --wrapper-prefix 'https://github.com/snakemake/snakemake-wrappers/raw/' --latency-wait 50 --scheduler 'ilp' --scheduler-solver-path '/home/uds_alma015/.conda/envs/abfe/bin' --default-resources 'mem_mb=max(2*input.size_mb, 1000)' 'disk_mb=max(2*input.size_mb, 1000)' 'tmpdir=system_tmpdir' --mode 2 && touch '/home/uds_alma015/GIT/BindFlow/ideas/reproduce error/.snakemake/tmp.lfq5jq4u/0.jobfinished' || (touch '/home/uds_alma015/GIT/BindFlow/ideas/reproduce error/.snakemake/tmp.lfq5jq4u/0.jobfailed'; exit 1)

P.s. ChatGPT goes in circle with this question :-)

Update

I even isolate more the error. The following other_job.sh (submitted to the cluster as sbatch other_job.sh) script also gave the same error:

#!/bin/bash
#SBATCH --partition=uds-hub 
#SBATCH --nodes=1
#SBATCH --cpus-per-task=12 
#SBATCH --mem=5000
#SBATCH --time=15:00
#SBATCH --job-name=reproduce_error
#SBATCH --output=reproduce_error.o
#SBATCH --error=reproduce_error.e

/home/uds_alma015/.conda/envs/abfe/bin/python3.9 -m snakemake --snakefile '/home/uds_alma015/GIT/BindFlow/ideas/reproduce_error/Snakefile' --target-jobs 'gmx:' --allowed-rules 'gmx' --cores 'all' --attempt 1 --force-use-threads  --resources 'mem_mb=1000' 'mem_mib=954' 'disk_mb=1000' 'disk_mib=954' --wait-for-files '/home/uds_alma015/GIT/BindFlow/ideas/reproduce_error/.snakemake/tmp.mqan6qbp' 'emin.tpr' --force --keep-target-files --keep-remote --max-inventory-time 0 --nocolor --notemp --no-hooks --nolock --ignore-incomplete --rerun-triggers 'mtime' 'software-env' 'params' 'code' 'input' --skip-script-cleanup  --conda-frontend 'mamba' --wrapper-prefix 'https://github.com/snakemake/snakemake-wrappers/raw/' --latency-wait 50 --scheduler 'ilp' --scheduler-solver-path '/home/uds_alma015/.conda/envs/abfe/bin' --default-resources 'mem_mb=max(2*input.size_mb, 1000)' 'disk_mb=max(2*input.size_mb, 1000)' 'tmpdir=system_tmpdir' --mode 2

And this is the command build by snakemake. It looks like that command does not interact (in some how) with the definitions of SBATCH. But still not sure.

Solution

  • I had a similar issue with snakemake, but with a different program. In the end I had to specifically unset some environment variables, because snakemake sets them for each instantiated shell. In particular OMP_NUM_THREADS was the root cause of my problem.

    Make sure to compare which environment variables are set within snakemake and in your regular script, this might give you a hint to find the culprit variable.