Skip to content

Using Containers on REPACSS

This guide covers using Apptainer/Singularity containers for running user-installed pre-built applications and complex software stacks on REPACSS.

🎯 When to Use Containers

Use containers when you need:

  • Pre-built applications that are difficult to install locally
  • Complex software stacks with many dependencies
  • Reproducible environments for research workflows
  • Software with specific system requirements
  • Applications that conflict with system libraries

🐳 Getting Started with Apptainer

Check Apptainer Availability

Apptainer is system-wide installed on REPACSS. Verify it's working:

repacss:$ apptainer --version
apptainer version 1.4.0-1.el9

No installation required - Apptainer is already available on all REPACSS nodes.

Basic Container Operations

Pull a Container

# Pull a specific application (example: Ollama)
apptainer pull ollama.sif docker://ollama/ollama:0.6.8

# Pull from a specific registry
apptainer pull myapp.sif docker://registry.example.com/myapp:latest

Run a Container

# Execute a command in a container
apptainer exec ollama.sif ollama list

# Run a container as an application
apptainer run ollama.sif

# Shell into a container
apptainer shell ollama.sif

📁 Working with Data

Binding Directories

Mount your data directories into the container:

# Bind a single directory
apptainer exec -B /path/to/data:/data myapp.sif bash

# Bind multiple directories
apptainer exec -B /path/to/data:/data,/path/to/results:/results myapp.sif bash

# Bind your home directory (common pattern)
apptainer exec -B $HOME myapp.sif bash

Example: AI Application Workflow

# Create a working directory
mkdir -p ~/ollama-work

# Run Ollama in container with data mounted
apptainer exec -B ~/ollama-work:/work ollama.sif ollama run llama3.1:8b

🚀 Running Containers in Jobs

Interactive Jobs

# Request an interactive session
interactive -p h100 -t 02:00:00 -g 1

# Load any required modules
module load cuda/12.6.2  # if using GPU

# Run your container
apptainer exec ollama.sif ollama run llama3.1:8b

Batch Jobs

Create a job script for batch execution:

#!/bin/bash
#SBATCH --job-name=ollama-job
#SBATCH --partition=h100
#SBATCH --cpus-per-task=4
#SBATCH --mem=8G
#SBATCH --gres=gpu:1
#SBATCH --time=02:00:00
#SBATCH --output=ollama_%j.out
#SBATCH --error=ollama_%j.err

# Load modules if needed
module load cuda/12.6.2

# Run container with data binding
apptainer exec -B $HOME:/home/user ollama.sif ollama run llama3.1:8b

🔍 Finding Containers

  • Docker Hub: docker://ollama/ollama:0.6.8
  • NGC (NVIDIA): docker://nvcr.io/nvidia/pytorch:23.12-py3
  • Biocontainers: docker://biocontainers/fastqc:v0.11.9_cv8

Scientific Computing Containers

# PyTorch with CUDA
apptainer pull docker://nvcr.io/nvidia/pytorch:23.12-py3

# TensorFlow
apptainer pull docker://tensorflow/tensorflow:latest-gpu

# Jupyter with scientific stack
apptainer pull docker://jupyter/datascience-notebook:latest

AI Applications

For trending AI applications like Ollama, see our dedicated guide:

🎯 Best Practices

Container Management

  • Use descriptive names for your containers
  • Version your containers (e.g., myapp_v1.0.sif)
  • Store containers in $WORK for large files
  • Document container contents and dependencies

Performance Considerations

  • Use local storage ($HOME or $WORK) for containers
  • Bind only necessary directories to avoid overhead
  • Consider container size for transfer and storage
  • Use appropriate partitions (GPU containers on h100 partition)
  • Use approved containers like Ollama for AI applications

Security

  • Only use trusted container sources (Docker Hub, NGC, etc.)
  • Verify container contents before running
  • Don't run containers as root unless necessary
  • Keep containers updated for security patches
  • Use approved applications like Ollama for AI workloads

🛠️ Troubleshooting

Common Issues

  1. Permission denied: Ensure container file is executable 

    chmod +x myapp.sif

  2. Cannot bind directory: Check directory permissions and existence 

    ls -la /path/to/data

  3. Container not found: Verify container file exists and path is correct 

    ls -la *.sif

  4. Out of disk space: Use $WORK directory for large containers 

    apptainer build $WORK/myapp.sif myapp.def

🆘 Support

For container-related issues or assistance, contact:

repacss.support@ttu.edu