code-server (VS Code in the browser)¶

code-server lets you run a full VS Code environment on a compute node and access it from any browser. No local installation needed, no extensions to configure on your machine. Just submit a job, open a URL, and start coding.

If you already use VS Code Remote SSH, think of code-server as the lighter alternative: everything happens in the browser, and any device with a web browser can be your workstation.

How it works¶

Since compute nodes are not directly accessible from outside the cluster, the setup involves three pieces:

1. A batch job that starts code-server on a compute node
2. An SSH tunnel from your local machine through the login node to the compute node
3. Your browser pointing to localhost

For more details on how SSH tunneling works, see the SSH tunneling section in the connection guide.

The following video walks through the entire process from start to finish:

Step 1: Submit the batch script¶

Save the following as code-server.sh:

#!/bin/bash
#SBATCH --job-name=code-server
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
#SBATCH --mem=16G
#SBATCH --time=8:00:00
#SBATCH --qos=regular
#SBATCH --output=code-server-%j.log

module load DIPC-Utils

start-code-server

The start-code-server script takes care of everything: it launches code-server, waits for it to be ready, and prints the connection information to the log file.

Submit it:

sbatch code-server.sh

If you are not familiar with the #SBATCH flags, the SLURM reference explains each directive in detail. Adjust --cpus-per-task, --mem, and --time to fit your needs.

start-code-server --port 9090 --password mysecretpass

Step 2: Check the log¶

Wait for the job to start running:

squeue -u $USER

Once the state changes to R (running), read the log file:

cat code-server-<JOBID>.log

You will see something like this:

========================================
 code-server is running
========================================

 Node:     hyperion-039
 Port:     8080

 1. SSH tunnel (run on your local machine):

    ssh -L 8080:hyperion-039:8080 <username>@hyperion.sw.ehu.es -N

 2. Open in your browser:

    http://localhost:8080

 3. Password: d51867d0ad62e2affc717da9

========================================

The log gives you everything you need: the tunnel command, the URL, and the password.

Step 3: Create the SSH tunnel¶

On your local machine, open a terminal and run the tunnel command from the log:

ssh -L 8080:hyperion-039:8080 <username>@hyperion.sw.ehu.es -N

This forwards your local port 8080 through the login node to port 8080 on the compute node. The terminal will appear to hang with no output; that is normal.

Step 4: Open your browser¶

Navigate to:

http://localhost:8080

Enter the password from step 2. You now have a full VS Code environment running on a compute node.

Working with your files¶

code-server starts in your home directory by default. To open a project on /scratch:

1. Click File > Open Folder
2. Navigate to /scratch/<username>/your-project
3. Click OK

Extensions¶

code-server uses the Open VSX registry instead of the Microsoft Marketplace. Most popular extensions (Python, Jupyter, GitLens, etc.) are available there.

To install an extension:

1. Open the Extensions panel from the left sidebar (or press Ctrl+Shift+X)
2. Search for the extension you want
3. Click Install

Extensions are installed on the cluster in ~/.local/share/code-server/extensions/ and persist across sessions, so you only need to install them once.

Running Jupyter notebooks¶

code-server can run Jupyter notebooks directly in the browser, just like the desktop version of VS Code.

Install the required extensions¶

From the Extensions panel (Ctrl+Shift+X), search and install:

1. Python
2. Jupyter

The following video shows how to install both extensions:

You only need to do this once; the extensions persist across sessions.

Register your conda environment as a kernel¶

code-server will not automatically detect conda environments. You need to register them as Jupyter kernels explicitly. This is a one-time step per environment.

Open the integrated terminal (Ctrl+`) and load the module:

module load Miniforge3

Then activate your environment and register it. The steps depend on whether your environment has a name or only a path prefix:

conda activate my_environment
python -m ipykernel install --user --name=my_environment --display-name "My Environment"

conda activate /scratch/$USER/env/my_project
python -m ipykernel install --user --name=my_project --display-name "My Project"

If ipykernel is not installed in your environment, add it first:

conda install ipykernel

See Adding custom kernels for more details on managing kernels.

Open a notebook and select the kernel¶

1. Open a .ipynb file from the file explorer (or create one with File > New File and save it as .ipynb)
2. In the top right corner of the notebook, click Select Kernel
3. Choose Jupyter Kernel
4. Select the kernel you just registered (e.g., "My Project")

You can now run cells and they will execute on the compute node using the packages from your selected environment.

Stopping the server¶

To end your code-server session, cancel the batch job:

scancel <JOBID>

You can find the job ID with squeue -u $USER.

Troubleshooting¶

See also¶