This script should be installed on keeling. It will start a jupyter lab session on a keeling compute node given inputted parameters, and then provide you a command to run in a separate terminal on your local machine to establish an ssh tunnel to a keeling compute node. It waits to print SSH commands + Jupyter URL until the compute node has the jupyter lab session ready.
git clone https://github.com/atmsillinois/jlab-slurm-screen
cd jlab-slurm-screen
chmod u+x jlab-slurm-screen.sh
You may wish to consider adding this folder to your path.
cd /path/to/jlab-slurm-screen/
./jlab-slurm-screen.sh hours=2 session=jlab cpus=4 mem=8G jlab_port=5678
If your session is called jlab (the default):
screen -S jlab -X quit
| Argument | Description |
|---|---|
| hours=2 | time of job in hours |
| session=jlab | prefix of slurm job |
| cpus=4 | number of cores |
| mem=8G | memory per core |
| partition=seseml | slurm partition |
| jlab_port=8888 | jupyerlab port to try (may change if not available) |
| dask_port=8787 | [optional] second port to tunnel (i.e., for dask dashboard) |
Behavior knobs (env vars)
- READY_WAIT=0 # seconds to wait here for NODE+PORT; 0 = wait indefinitely (default)
- PRINT_POLL=1 # polling interval (seconds) while waiting; default 1
- BASTION= # optional ProxyJump host, e.g. bastion.illinois.edu (adds -J to OUTER ssh)
- The session should be persistent for the alotted time. Thus, to reconnect if you lose your connection to keeling during the alotted time, simply re-establish the ssh tunnel.
- Debugging info is available in
~/.jlab/<session>_<stamp>.{log,url,port,node,where,tunnel}on keeling.