Skip to main content

Source repo: sdsc-summer-institute-2025 | Branch: main | Last synced: 2026-04-24 10:27:17.425 UTC

Using Singularity to run a Docker container on Expanse

This guide explains how to pull the jupyter/datascience-notebook container from Docker Hub using Singularity on SDSC Expanse, and then how to launch it using Galyleo.

1. Launch an Interactive Debug Job​

Use srun to get access to a compute node with local SSD for a temporary build environment, during the Summer Institute:

srun-compute

afterwards:

srun \
  --partition=debug \
  --nodes=1 \
  --ntasks=1 \
  --cpus-per-task=4 \
  --mem=8G \
  --time=00:30:00 \
  --pty bash

2. Load SingularityPro​

module load singularitypro

3. Set Environment Variables for Singularity​

Use the local SSD job space for caching and temporary build files:

export SINGULARITY_CACHEDIR="$SLURM_TMPDIR/singularity_cache"
export SINGULARITY_TMPDIR="$SLURM_TMPDIR"
mkdir -p "$SINGULARITY_CACHEDIR"

4. Pull the Docker Image and Save to Lustre Scratch​

Use the temp_project subdirectory in your Lustre scratch for storing the final .sif:

mkdir -p /expanse/lustre/scratch/$USER/temp_project

singularity pull /expanse/lustre/scratch/$USER/temp_project/datascience-notebook_latest.sif \
    docker://jupyter/datascience-notebook:latest

This command converts the Docker image to a Singularity .sif file and stores it in your persistent Lustre scratch space.

5. Launch the Container with Galyleo​

You can now launch the .sif container using Galyleo:

bash launch_galyleo_singularity.sh

📌 Make sure to update the path to your .sif file if you stored it in a different subfolder.

This will launch a Jupyter Notebook environment based on the jupyter/datascience-notebook container using Galyleo.

Notes​

  • Avoid building Singularity images on the login node — always use a compute node.
  • $SLURM_TMPDIR is local SSD on compute nodes and ideal for temporary build/cache use.
  • /expanse/lustre/scratch/$USER/temp_project is recommended for storing large container files.
  • Refer to the official docs at hpc-training.sdsc.edu/galyleo for full CLI options and examples.
  • 🚫 The jupyter/datascience-notebook container does not support GPU acceleration out of the box, for that you need to customize a naked-singularity container