Visual Studio Code on HPC Clusters

VS Code is a very popular IDE with a great Julia extension (website). Running it on a cluster typically isn't straightforward and we want to highlight the most important steps here.

Content

What does "Running VS Code on a cluster" actually mean?

To be clear: What it does not mean is to have VS Code running locally and using its terminal (at the bottom) the ssh to the cluster and running julia on it.

Instead, we want to connect VS Code to a little server component that will run on the cluster. Hence, only the GUI will run locally and everything else will run on the cluster. This includes VS Code extensions, like the Julia extension and will eventually give us an integrated REPL, a debugger, a plot pane, etc., all working remotely on the cluster.

Running VS Code on a login node (SSH)

Note: We assume that the hostname of the cluster is mycluster.

1. Make sure that you have password-less SSH access to the cluster (ssh mycluster should bring you to the cluster).

2. In VS Code, press F1 and run the Remote-SSH: Open SSH Host... command.

   • If the command isn't available, make sure that Remote - SSH extension is installed.

3. Enter mycluster and press enter.

After a second or two, you should have VS Code running on a login node! 🎉

⤴ back to Content

Installing the Julia extension

1. Open the extensions view (press CTRL/CMD + SHIFT + X).

2. Search for julia.

3. Click on install.

Making the Julia extension work

Typically, the julia command is not directly available on a cluster but one must load a Julia module first (e.g. module load julia). Unfortunately, the Julia extension doesn't know about this, so we need to teach it.

1. Prepare a Julia wrapper script (julia_wrapper.sh) like the following:

#!/bin/bash

# Making module command available
# ------------------------------------------------------------
export MODULEPATH=/etc/modulefiles:/usr/share/modulefiles
export LMOD_SYSTEM_DEFAULT_MODULES="DefaultModules"
source /usr/share/lmod/lmod/init/profile
module --initial_load restore
# ------------------------------------------------------------

# Load julia
module load julia

# Pass on all arguments to julia
exec julia "${@}"

Note the module name as well as (at least) the first two lines of the block that makes the module command available are system specific and must be adjusted. To figure our the latter, you may run the following in a shell on the cluster: echo $MODULEPATH and echo $LMOD_SYSTEM_DEFAULT_MODULES.

2. Make the wrapper script executable: chmod +x julia_wrapper.sh

3. Check that it works: ./julia_wrapper.sh should open the Julia REPL.

Once the wrapper is ready and functional, we need to point the Julia extension to it.

4. Open the VS Code Settings (press CTRL/CMD + ,).

5. Click on the tab Remote [SSH: mycluster].

6. Search for Julia executable.

7. Insert /path/to/julia_wrapper.sh into the text field under Julia: Executable Path.

If ALT/OPTION + J followed by ALT/OPTION + O (or pressing F1 and executing the Julia: Start REPL command) successfully spins up the integrated Julia REPL, you know that the setup is working! 🎉

Important note: If you want all the nice features of the Julia extension - like inline evaluation, plotting, debugging, etc. - you need to use the integrated REPL started by the shortcuts mentioned above. If you just type julia in a terminal (at the bottom), the resulting REPL will not be connected to the extension!

⤴ back to Content

Can I run VS Code on a compute node? (SSH)

Maybe. The first thing you need to check is whether it is at all possible for you to connect to a compute node via SSH, as not all HPC centers allow this. (Note that typically you can only ssh to compute nodes that are part of a job that you have currently running.)

Given that you can ssh to compute nodes, what you want to do next is add a generic (wildcard) entry, that covers all compute nodes, to your .ssh/config. Assuming that the hostnames of the compute nodes start with cn (e.g. cn1201 and cn2134 etc.) and assuming that you already have a ssh config entry for mycluster, this could look like this:

Host cn*
    HostName %h
    ProxyJump mycluster
    User [USERNAME]
    IdentityFile [PATH TO PRIVATE KEY]
    IdentitiesOnly yes

Afterwards, you can use the following steps to run VS Code on a compute node:

1. Open a terminal (outside of VS Code)

2. Login to the cluster (ssh mycluster)

3. Get an interactive session on a compute node (and write down the name of the node that has been assigend to you, say, cn1201).

4. Follow the steps under Running VS Code on a login node (SSH) above but replace the hostname mycluster by the name of the compute node (e.g. cn1201).

⤴ back to Content

Running VS Code remotely via "remote tunnels"

Microsoft provides "Remote tunnels" as an alternative to SSH to connect VS Code to remote systems.

Pro

• Allows you to connect to any machine with internet access (including cluster compute nodes that you can't ssh to!).

Con

• Requires a GitHub / Microsoft account.

• Microsoft serves as the tunnel provider (man in the middle).

How to do it?

1. Download the VS Code command line interface from https://code.visualstudio.com/Download (search for CLI).

2. Copy the single code binary to the cluster (and put the directory on PATH).

3. Run code tunnel --verbose and follow the steps (to connect to your GitHub account).

4. Afterwards, open VS Code and install the Remote - Tunnels extension.

5. Run the command Remote-Tunnels: Connect to Tunnel... and connect to the remote node.

⤴ back to Content