Before you start#
Euler has login nodes and compute nodes. Prepare files and control jobs from the login node. Slurm is the scheduler; it runs actual computation on a compute node. A batch job is a submitted script, not merely a file in a folder. sbatch submits one batch script and returns its job ID. Keep that number in $HOME, your small private Euler folder, and use the same numeric job ID later.
| Stage | What you inspect | What it establishes |
|---|---|---|
| Create and inspect | Script text and syntax result | The file is ready, not running |
| Submit once | Returned numeric ID | Slurm accepted that submission |
| Observe | squeue for that ID |
Whether it is currently queued or running |
| Read the result | sacct, seff and its output |
Recorded state, resources and expected calculation |
GiB is the memory unit used here. The teaching request is one CPU, 1 GiB per CPU, two minutes, no GPU and the already verified Euler Python environment. This is a workflow test, not a resource recommendation for a research program.
If the script prints correctly but the queue is empty, what has actually happened so far? Decide which stage you reached before trying another command. The recipe follows one job throughout; you do not need a second submission.
Read the completed job without mixing its rows
| Form observation | Where to compare it |
|---|---|
| Account | Main row: normal/es_fuge or es_fuge; normal/es+ is truncated |
| State and exit code | Main row: COMPLETED and 0:0 |
| CPU and requested memory | Main row: one CPU and 1G (per-CPU suffix may appear) |
| Observed memory | Relevant .batch step; not the requested value |
| Program output | Same ID: python_environment=passport-python and 5 squared is 25 |
Read seff for that same job. Keep actual logs and identifiers private; enter only the requested observations in the local form.
Submitting again because a short job disappeared from the queue. Keep the stored ID and read its accounting and logs; queue visibility is not its outcome.