Attend gray_scott_school school with Juliet

These pages are here to help you use the Juliet cluster for the gray_scott_school school. If you run into any trouble, do not hesitate to ask questions on school #romeo-support channel on the official Gray Scott Schoot server.

Connect to juliet

You have received a login and password by e-mail. You may connect to juliet using these logins and password by typing in a command prompt (ex: PowerShell for windows)

ssh username@juliet.mesonet.fr

You should not be prompted for a password for juliet (you might still be prompted for a password if your ssh key is password protected). If you are, try to specifcy the type of ssh key with

ssh -i path/to/your/public/ssh_key username@juliet.mesonet.fr

If you are still prompted for a password, check if your ssh key was correctly registered or contact support.

Access your classes

Most of the classes use apptainer containers. These containers are on the directory /projects/m26116/containers/

You can also access specific instructions for each class here

Connect to a compute node

Do not compute on juliet1

juliet1 on which you first connect is a login node, which is shared between all users and not made to run computations.

To be able to run the codes you must be on a compute node which is either juliet2 juliet3 or juliet4. In case of a doubt, type the command

hostname

to know where you are.

Default process to access compute nodes

To connect to a compute node you will need to use slurm

By default you can run the command:

srun --reservation=gray_scott_school --partition=mesonet --account=m26116 --time=10:00:00 --mem=64G -c 16 --pty /bin/bash

If you need a GPU, run the command

srun --reservation=gray_scott_school  --partition=mesonet --account=m26116 --time=10:00:00 -c 16 --mem=64G --gres=gpu:1 --pty /bin/bash

Both commands will open a terminal on a compute node.

Avoid GPU issues

Due to a technical issue, for your job to properly take GPUs into account you will need to enter this command as soon as you connect to a compute node

 [[ -n "$SLURM_STEP_GPUS" ]] && export CUDA_VISIBLE_DEVICES=$SLURM_STEP_GPUS || export CUDA_VISIBLE_DEVICES=$SLURM_JOB_GPUS

For podman users

While we mainly use apptainer, you can use podman on every container of the school if you know how to use it. You will need the adresse on the Container List time table

With a few exceptions containers have not been tested with podman so only choose this option if you know what you are doing.

If you want to use podman use salloc instead of srun

salloc  --reservation=gray_scott_school  --partition=mesonet --account=m26116 --time=10:00:00 -c 16 --mem=64G

OR

salloc --reservation=gray_scott_school  --partition=mesonet --account=m26116 --time=10:00:00 --mem=64G --gres=gpu:1

then

ssh julietX

where julietX can be juliet2, juliet3 or juliet4

Run the containers

To run the containers you can run the command

apptainer shell path/to/container.sif

Alternatively, you can run a bash script that will do it for you. Bash scripts for classes are available at

bash /projects/m26116/scripts/script.sh

and have the same name as the container with .sh instead of .sif

** /!\ MAKE SURE YOU ARE CONNECTED TO A COMPUTE NODE BEFORE STARTING A CONTAINER OR A SCRIPT /!\ **

Get the files from the container

If you use the script, you will need to copy the files from their location on the container to your home directory, otherwise you will not be able to change them.

If you want to edit the files in the container directly, you can add the --writeable-tmpfs option to the apptainer shell command.