Running SnakeMake on cluster

Question

I do not understand how to specify correctly the parameters on a SLURM cluster for snakemake to use them. I tried submitting the following SLURM file, but it does not work this way and the number of cores used is only 1, not 20:

#!/bin/bash

#SBATCH -p standard
#SBATCH -A overall 
#SBATCH --time=12:00:00
#SBATCH --output=snakemake%A.out
#SBATCH --error=snakemake%A.err
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=20
#SBATCH --mem=120000

snakemake

Then, I tried following snakemake tutorial. And created cluster.json based on the SLURM parameters that I need:

{
    "__default__" :
    {
        "A" : "overall",
        "time" : "24:00:00",
        "nodes": 1,
        "ntasks": 1,
        "cpus" : 20,
        "p" : "standard",
        "mem": 120000,
        "output": "snakemake%A.out",
        "error": "snakemake%A.err"
    }
}

And ran snakemake inside a newly created snakemake.sh script:

#!/bin/bash

snakemake -j 999 --cluster-config cluster.json --cluster "sbatch -A 
{cluster.A} -p {cluster.p} \ 
-t {cluster.time} --output {cluster.output} --error {cluster.error} -- 
nodes {cluster.nodes} \
--ntasks {cluster.ntasks} --cpus-per-task {cluster.cpus} --mem 
{cluster.mem}"

And it is giving me an error now:

sbatch: error: Unable to open file
/bin/sh: line 1: -t: command not found Error submitting jobscript (exit code 127):

I am now completely lost at what I shoould actually do. I would prefer plain regular .slurm file submission, but how to make snakemake use them? Any suggestions would be greatly appreciated.

I removed \ - line separators in the snakemake.sh script:

#!/bin/bash

snakemake -j 10 --cluster-config cluster.json --cluster "sbatch -A {cluster.A} -p {cluster.p} -t {cluster.time} --output {cluster.output} --error {cluster.error} --nodes {cluster.nodes} --ntasks {cluster.ntasks} --cpus-per-task {cluster.cpus} --mem {cluster.mem}"

And it started to run. It is not convenient though for me. I would rather prefer submitting just one job using .slurm file passing all of the parameters from #SBATCH. Is it possible?

Answer 1

You can omit --nodes, you need the following:

#!/bin/bash
#SBATCH --ntasks-per-node=1
#SBATCH -c threads_from_snakemake
#SBATCH -p some_partition

Commands go here

For slurm, you may want to modify my SlurmEasy script. We use that with snakemake all the time, usually of the form:

snakemake -c "SlurmEasy -t {threads} -n {rule} --mem-per-cpu {cluster.memory}" ...

You'll also want units with your memory requests, like 12G. Also, in general it's best not to submit snakemake itself to the cluster, rather run it on an interactive node and have it submit jobs on your behalf (if the admins complain point out that snakemake is barely using any resources on the head node).

Answer 2

While I agree with not submitting snakemake to the cluster, it's not a great solution to make your pilot program run interactively - automation via batch script should not be dependent on an open shell.

How about allocating a whole node with many cores and working within the GNU parallel abilities of snakemake.