Scientific software with EESSI
EESSI (the European Environment for Scientific Software Installations, pronounced "easy") is a shared, read-only collection of thousands of scientific applications, built and optimised by a collaboration of European HPC centres. On the DSRI it is mounted straight into your pod, so the software is simply there: you load what you need with module load and start working.
The software is compiled for several CPU architectures, and the right version for the node your pod lands on is selected automatically. You are not running a generic build.
There are two ways to use it:
- The EESSI Terminal — a shell in your browser, for exploring, testing commands and inspecting results.
- A batch job — for analyses that run for hours and should not depend on your laptop staying awake.
Best practice on how to use it would be to work out the command in the terminal, then submit the real workload as a job.
The EESSI Terminal
Deploy
Search the catalogue in the web UI for EESSI Terminal and instantiate the template.
Opening the terminal
Get the route URL via the CLI:
oc get route eessi-terminal -o jsonpath='{.spec.host}{"\n"}'
You can also choose to navigate to the route URL via the DSRI console.
What is in the pod
| Path | What it is |
|---|---|
/data/persistent | Your home directory. Lives on CephFS and survives restarts. |
/cvmfs/software.eessi.io | The EESSI software stack. Read-only. |
Everything you want to keep must be under /home/persistent. Files written anywhere else are lost when the pod restarts.
The terminal is for interactive work. Anything that runs for more than a few minutes belongs in a job, which survives pod restarts and node maintenance. See running a batch job below.
Finding and loading software
EESSI is already initialised when your shell opens. To see what is available:
module avail
The first time you run this it is slow, because the software catalogue is being fetched and cached on the node. Afterwards it is fast.
The list is long, so search for what you need. For example, search for the gzip module:
module avail gzip
Load for example the GCC module, and test it:
module load GCC
gcc --version | head -1
Some other useful commands:
module list # what you have loaded right now
module unload GCC # unload specific module
module purge # unload everything
module load only affects your current shell session. It is not remembered after a restart, and it does not carry over into a job. Put your module load lines in your script.
Your session survives a closed browser
The terminal runs inside a tmux session, so closing the tab, dropping off VPN or closing your laptop does not kill what you are running. When you open the terminal again you are reattached to the same session, with your command still running.
The usual tmux keys work:
Ctrl-b c new window
Ctrl-b n next window
Ctrl-b [ scroll back through output (q to leave scroll mode)
tmux protects you against losing your connection. It does not protect you against the pod being restarted or the node being drained for maintenance, which can happen at any time. Work that takes hours belongs in a job.
Running a batch job
A job runs without anyone connected to it. It gets its own resources, keeps its logs, and is removed automatically when it finishes. This is the right way to run anything substantial.
For more information on how to run jobs on the DSRI please refer to our guidelines for running jobs.
Below is a complete job you can copy and adapt. Save it as eessi-job.yaml.
# Persistent directory for your data and results. It stays when the job is
# deleted. If you already have a workspace, remove this part and point
# claimName further down at your own PVC.
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: eessi-job-pvc
spec:
accessModes: [ReadWriteMany]
storageClassName: ocs-storagecluster-cephfs
resources:
requests:
storage: 10Gi
---
apiVersion: batch/v1
kind: Job
metadata:
name: eessi-job
spec:
backoffLimit: 1
ttlSecondsAfterFinished: 86400
template:
spec:
restartPolicy: Never
containers:
- name: job
image: ghcr.io/maastrichtu-library/eessi-terminal:latest
command: ["/bin/bash", "-c"]
args:
- |
set -eo pipefail
# unset 'u' flag as it results in abortion of the script due to EESSI's init script.
# only set it after initializing EESSI.
set +u
source /cvmfs/software.eessi.io/versions/2026.06/init/bash
set -u
# ----- your work starts here -----------------------------
echo "Loading module GCC..."
module load GCC
module list
echo "Creating directory /data/persistent/my-analysis"
mkdir -p "$HOME/my-analysis"
cd "$HOME/my-analysis"
echo "Saving results..."
echo "$(date -Is) $(gcc --version | head -1)" >> results.txt
echo "Finished saving results..."
# ----- your work ends here -------------------------------
env:
- name: HOME
value: /data/persistent
- name: OMP_NUM_THREADS
value: "8"
resources:
requests: {cpu: "8", memory: 16Gi, ephemeral-storage: 1Gi}
limits: {cpu: "8", memory: 16Gi, ephemeral-storage: 4Gi}
securityContext:
allowPrivilegeEscalation: false
capabilities:
drop: [ALL]
volumeMounts:
- name: cvmfs
mountPath: /cvmfs/software.eessi.io
mountPropagation: HostToContainer
readOnly: true
- name: data
mountPath: /data/persistent
volumes:
- name: cvmfs
ephemeral:
volumeClaimTemplate:
spec:
accessModes: [ReadOnlyMany]
storageClassName: software-eessi-io
resources:
requests:
storage: 1Gi
- name: data
persistentVolumeClaim:
claimName: eessi-job-pvc
What to change
Three things, in practice.
Your script. Everything between the two comment lines. Note that the job starts with a clean shell, so the module load lines you used interactively have to be repeated here.
The resources. requests is what the scheduler reserves for you and limits is the ceiling. Exceeding the memory limit kills the job, so give it headroom. Keep OMP_NUM_THREADS equal to the CPU limit, otherwise your software will try to use more cores than it has been given and run slower, not faster.
The workspace. If you also run the EESSI Terminal, delete the PVC part at the top and set claimName to your terminal's home PVC (eessi-terminal-home by default). The job then reads and writes the same files you see in the terminal.
The parts to leave alone
mountPath: /cvmfs/software.eessi.io
The software stack must be mounted at exactly this path. EESSI's internal references are absolute, so mounting it anywhere else produces a confusing EESSI repository ... not found! error even though the files are clearly there.
- name: cvmfs
ephemeral:
volumeClaimTemplate:
This is how the pod gets access to EESSI. It is created with the pod and removed with it, so submitting the same job repeatedly leaves nothing behind. Nothing is stored in it: the requested 1Gi is ignored and is only there to satisfy project policies.
Submitting and following along
oc apply -f eessi-job.yaml
Watch it:
oc get job eessi-job -w
Follow the output:
oc logs -f job/eessi-job
To run it again, delete the old job first. The name has to be free:
oc delete job eessi-job
oc apply -f eessi-job.yaml
Your results stay in the workspace PVC, so deleting the job does not delete your output.
Using a GPU
Find out how to enable a GPU in your project.
CUDA-enabled modules in EESSI are marked with (gpu) in the output of module avail.
Running the same analysis many times
If you need to run something tens or hundreds of times, do not submit tens or hundreds of individual jobs. Use an indexed job, which packs several runs into each pod. See Guidelines for running jobs for how to set that up; combine it with the EESSI volume and image shown above.
Troubleshooting
EESSI repository ... not found!
The software stack is mounted at the wrong path. Check that mountPath is exactly /cvmfs/software.eessi.io.
module: command not found
The initialisation line is missing from your script, or it ran before the mount was ready. Make sure your script starts with:
source /cvmfs/software.eessi.io/versions/2026.06/init/bash
cvmfs transport endpoint is not connected
The mount was interrupted, usually after cluster maintenance. Restart your deployment, or resubmit your job:
oc rollout restart deployment/eessi-terminal
Everything is slow the first time
Expected. Software is fetched and cached on the node on first use. The second run of the same application is much faster.
My software is not in EESSI
Check the list of available software first, since the names are not always what you expect. If it genuinely is not there, you can still install your own software into your workspace on top of EESSI, or contact us.
Getting help
If you have questions about EESSI on the DSRI contact the RCS team.
Questions about the software stack itself, or a request to have an application added: see EESSI support.