| tags |
|---|
farm, jupyter |
[toc]
Titus Brown, [email protected]
June 19, 2023
To edit: / latest on github
(Adapted from Nick Ulle's tutorial!)
The below uses the mambaforge installation of conda.
First, create a mamba environment with jupyter notebook installed:
mamba create -y -n jup notebook nb_conda_kernels
Activate it:
mamba activate jup
Set a password:
jupyter notebook password
Create a script run-jupyter-on-node.sh containing:
#! /bin/bash -eu
# initialize conda
conda_base=$(conda info --base)
. ${conda_base}/etc/profile.d/conda.sh
# activate relevant environment - CUSTOMIZE ME
conda activate jup
### pick a port
read LOWERPORT UPPERPORT < /proc/sys/net/ipv4/ip_local_port_range
port="$(shuf -i $LOWERPORT-$UPPERPORT -n 1)"
while :; do
netstat -taln | egrep ":$port.*LISTEN" || break
port="$(shuf -i $LOWERPORT-$UPPERPORT -n 1)"
done
### print out stuff
cat <<EOF
You need to use SSH tunneling to port forward from your desktop to Jupyter Lab.
Run the following command in a new terminal on your machine:
ssh -L${port}:${HOSTNAME}:${port} [email protected] -N
Then point your browser to http://localhost:${port}
EOF
### run me!
jupyter notebook --no-browser --ip=${HOSTNAME} --port=${port}
and then run with srun:
srun -p high2 --time=3:00:00 --nodes=1 \
--cpus-per-task 1 --mem 5GB \
--pty bash ./run-jupyter-on-node.sh