Slurm Snakemake Profile

This is a snakemake profile that you can use to have any snakemake file submit jobs on Sherlock using slurm.

Usage

First, download this repo and install the profile by running

bash install_snakemake_profile.sh

To use the profile, just run

snakemake --profile slurm

You can add any additional arguments to snakemake just like normal.

Notes on Snakemake Cluster Mode

• In cluster mode, quitting snakemake will not cancel jobs that have already been submitted. Use scancel to clean up any remaining snakemake jobs before restarting snakemake.
• The snakemake log is decidedly less helpful in saying why a rule crashed in cluster mode. See the Accessing Logs section below for tips.

Customizing job resource requests

If you have a snakemake rule that needs more time, memory, or cores than the defaults listed in cluster.json, you can add parameters to your snakemake rule to specify the following:

• CPUs: Set the threads attribute as normal in snakemake
• Memory, time, partition, priority: Set values for cluster_memory, cluster_time, cluster_partition, or cluster_priority under the params section of the rule. These correspond to the --mem, -t, -p, and --qos arguments to sbatch. For example, to make a job with 8 cores, 12G of memory and 4 hours of time do:

rule my_job:
    input: ...
    output: ...
    threads: 8
    params:
        cluster_memory = "12G",
        cluster_time = "4:00:00"
    shell: ...

Accessing logs

Log records containing the outputs of all submitted slurm jobs are saved in your snakemake working directory as follows:

• Standard out: .slurm-logs/{rule_name}/snake.{wildcard_values}.{jobid}.out
• Standard error: .slurm-logs/{rule_name}/snake.{wildcard_values}.{jobid}.err

Usually, I end up wanting to look at logs from a specific jobid that I know crashed. In this case it is tedious to type out the rule name and wildcard values. Instead, for example if job 54085154 crashed I will run:

less .slurm-logs/*/*54085154.out
less .slurm-logs/*/*54085154.err

Credits

The submission scripts were adapted from: https://github.com/Snakemake-Profiles/slurm/