How to set binding memory limits in snakemake

Question

I would like to have Snakemake set binding memory limits for individual rules. Based on the snakemake documentation, mem_mb argument seems like it would work, but the job uses more memory than I've allocated.

Here's a simple rule that uses several GB of memory. I would like the rule to be stopped once it hits the memory limit, but it completes without issue.

rule:
    output:
        "a"
    threads: 1
    resources:
        mem_mb = 100
    shell:
        """
        python3 -c 'import numpy; x=numpy.ones(1_000_000_000)'
        touch a
        """

Is it possible to make this limit bind? I'd like a solution that's portable, working for Windows and Linux. I'm using snakemake locally, not with a batch scheduler or container setup.

Answer 1

I have absolutely no experience with this, so I can't really say if this is recommended or works well across platforms, but this seems to work on my computer (Ubuntu):

rule all:
    input:
        "a"

rule:
    output:
        "a"
    threads: 1
    resources:
        mem_mb = 100
    params:
        max_mem=lambda wildcards, resources: resources.mem_mb * 1024
    shell:
        """
        ulimit -v {params.max_mem}
        python3 -c 'import numpy; x=numpy.ones(1_000_000_000)'
        touch a
        """

See here for more info.

I don't believe SnakeMake offers an "out of the box" solution for this.

Answer 2

I don't have much experience with this myself but note that the use of the resources directive should be accompanied by the command line option --resources. In your case, you could execute:

snakemake -j 10 --resources mem_mb=500

and this will ensure that jobs running at the same time will not exceed mem_mb = 500 (and at most 10 jobs run at the same time). However, a rule with more than mem_mb=500 will still run as a single job. To prevent that I think Maarten-vd-Sande's solution is the best I can think of.