Skip to content

Latest commit

 

History

History
140 lines (85 loc) · 4.67 KB

README.md

File metadata and controls

140 lines (85 loc) · 4.67 KB
Raw

Using Snakemake on Aalto Triton: Parameter Grid Example

In this example Snakemake workflow, we loop over two parameters and run a Python script using each parameter combination as input in parallel.

The workflow can be cloned and run on the Aalto Triton cluster following the instructions below.

You can use it as a starting point for your own workflow.

Setup and Run Snakemake on Aalto Triton

Instructions to setup and run Snakemake on Aalto Triton.

Setup

First, connect to Triton using SSH.

Change directory to your work directory

cd $WRKDIR

Clone this repository and change directory

git clone https://github.com/AaltoRSE/snakemake-triton-example.git
cd snakemake-triton-example/parameter-grid

Load mamba tool from a module and check that it works by printing its version number

module load mamba
mamba --version

Use mamba to create a conda environment containing Snakemake and the Slurm executor plugin

mamba env create --file snakemake.yml --prefix env/

Run

Submit the Slurm job which starts Snakemake with

sbatch snakemake_slurm.sh

The batch script snakemake_slurm.sh requests resources to run Snakemake itself.

We start Snakemake via a batch script so that it runs non-interactively and we are free to close our connection to Triton while we are waiting for the actual workflow computations to finish.

The contents of the batch script:

#!/bin/bash
#SBATCH --job-name=snakemake_slurm
#SBATCH --time=01:00:00            # Runtime needs to be longer than the time it takes to complete the workflow. Modify accordingly!
#SBATCH --cpus-per-task=2          # Generally enough CPUs to run Snakemake. No need to modify.
#SBATCH --mem=2G                   # Generally enough RAM to run Snakemake. Modify if Snakemake process runs out of memory.
#SBATCH --output=snakemake_slurm-%j.out
#SBATCH --error=snakemake_slurm-%j.err

module load mamba
source activate env/
snakemake --snakefile workflow/Snakefile --profile profiles/slurm/ --software-deployment-method conda --conda-frontend mamba --cores 2

Here's a breakdown of the Snakemake command (the final line):

  1. --snakefile workflow/Snakefile

    Explicitly state which Snakefile to run.

    If omitted, Snakemake will try to find a Snakefile in the project root folder or under the workflow/ folder.

    Check workflow/Snakefile for details of the workflow logic.

  2. --profile profiles/slurm/

    Explicitly state the folder from which Snakemake looks for the config.yaml file.

    The config.yaml defines the cluster resources to request for rules in the Snakefile (cluster partition, number of CPUs and GPUs, RAM, runtime).

    Check profiles/slurm/config.yaml for details.

  3. --software-deployment-method conda

    Tell Snakemake to create and use the rule-specific conda environments defined in the Snakefile.

  4. --conda-frontend mamba

    Use mamba instead of conda.

  5. --cores 2

    Tell Snakemake to use two CPUs to run itself.

The intermediate and final result files will be written to the project root folder under results/.

How to adapt the project for your own workflow

In order to adapt the project to your own workflow, do the following:

  1. Replace the data and scripts in folders resources/ and workflow/scripts with your own, respectively.
  2. Replace/modify the workflow in workflow/Snakefile with your own.
  3. Add/modify the requested computational resources for the Snakefile jobs (rules) to the profile file in profiles/slurm/config.yaml.
  4. Increase the runtime value in the batch script snakemake_slurm.sh so that it is longer than the time it takes to complete the workflow. (Start e.g. with 12 hours and increase if necessary.)

Setup and Run Snakemake locally

Finally, it is often convenient to develop the workflow locally on your own laptop/desktop computer before moving onto Triton to run actual computation-intensive experiments.

To run the workflow locally, first clone the repository

git clone [email protected]:AaltoRSE/snakemake-triton-example.git
cd snakemake-triton-example/parameter-grid

Create the conda environment and activate it

mamba env create --file snakemake.yml --prefix env/
mamba activate env/

Then, run Snakemake with

snakemake --snakefile workflow/Snakefile --software-deployment-method conda --conda-frontend mamba --cores 1

Since we now omitted the --profile option which defined the Slurm resource configuration, all the computations will be run locally instead.