Run WRF with sbatch on Sapelo2¶

Use this guide after WPS has already created met_em.* files for your case.

The repository includes:

examples/sapelo2-single-node.sbatch

It requests one node and 16 MPI tasks.

You do not need an interactive shell to submit an sbatch job.

cd ~/work/project/wrfkit

WRFKIT_CASE=athens-minimal \
  sbatch examples/sapelo2-single-node.sbatch

Slurm prints a job number:

Submitted batch job 48718022

Your number will be different.

2. Check the queue¶

squeue -j YOUR_JOB_ID

Typical states:

State	Meaning
`PD`	Waiting in the queue
`R`	Running
gone from `squeue`	Finished or stopped; check `sacct`

3. Check the final Slurm result¶

sacct -j YOUR_JOB_ID \
  --format=JobID,State,ExitCode,Elapsed,NodeList,AllocCPUS

A normal run should show the main job as:

COMPLETED   0:0

4. Check WRF itself¶

CASE=athens-minimal

LOG=$(find ".wrfkit/logs/$CASE" \
  -maxdepth 1 -type d -name '*_wrf' | sort | tail -1)

tar -xOf "$LOG/native-logs.tar" rsl.out.0000 |
  grep "SUCCESS COMPLETE WRF"

What wrfkit does with your 16 tasks¶

You request resources from Slurm:

1 node
16 tasks / CPUs for WRF

wrfkit then uses the validated single-node bridge:

%%{init: {"flowchart": {"nodeSpacing": 20, "rankSpacing": 30}}}%%
flowchart LR
    A["Slurm allocation<br/>1 node · 16 CPUs"]
    B["Child srun step<br/>1 task · 16 CPUs"]
    C["Enter rootless Nix once"]
    D["Pinned OpenMPI<br/>mpirun -np 16"]
    E["16 WRF MPI ranks"]

    A --> B --> C --> D --> E

This avoids entering the rootless-Nix environment once per MPI rank.

Do not wrap wrfctl in another 16-rank srun

Avoid:

srun -n 16 ./wrfctl exec wrf --case my-case

Use the command already inside the example script:

./wrfctl exec wrf --case my-case

Use another case¶

WRFKIT_CASE=my-case \
  sbatch examples/sapelo2-single-node.sbatch

The example assumes met_em.* already exists for that case. It runs prepare wrf, real, and wrf.

Multi-node execution is not yet claimed as validated.