Skip to content

Configuring Python for scode

This companion guide explains how to get a rock‑solid Python workflow inside a VS Code Web session launched with scode. You’ll learn:

  1. Which VS Code extensions to install (and how to install them in one shot).
  2. How to load Python (and any supporting toolchains like CUDA or Java) before the server starts so they’re visible to both VS Code integrated terminals and Jupyter kernels.
  3. How the VS Code Python extension discovers interpreters, and what that means for:
    • plain .py files, and
    • Jupyter notebooks.
  4. Tips for refreshing kernels and troubleshooting interpreter‑detection quirks.

🚀 Quick Start

# On the login node
module load scode   # activate scode
scode ext install ms-python.python ms-toolsai.jupyter # Install essential extensions

# Launch a 2‑hour session with 16 GB RAM and pre‑loaded Anaconda
# Remove Anaconda module loading if you already have it loaded in your ~/.bashrc
scode serve-web \
  --setup-command "module load python/anaconda-2023.09 && source activate base" \
  -- --account <pi-account> --time 02:00:00 --mem 16G

1  Installing Python Extensions

Run this on the login node after module load scode but before launching the server.

scode ext install ms-python.python ms-toolsai.jupyter

These extensions provide essential support for Python development and interactive notebooks:

  • ms-python.python adds language features like IntelliSense, linting, debugging, and environment management for Python.
  • ms-toolsai.jupyter enables support for running Jupyter notebooks within the editor, including code execution and rich outputs.

More extensions be found on the VS Code Extension Marketplace.

Extensions are cached in ~/.scode/envs/stable/<env>/extensions/ by default, so subsequent sessions will start with the extensions you have previously installed.

Make sure to periodically run scode ext update to keep your extensions up to date.

2  Environment Setup & Tips

Jupyter kernels started within VS Code Web inherits the shell environment of the VS Code Server. So it is recommended to have the Python environment configured before starting the VS Code server with scode.

Two reliable patterns keep Python (and friends like CUDA or Java) on the PATH for both VS Code integrated terminals and Jupyter kernels:

Add your module‑load and activation commands to ~/.bashrc so they run for every interactive shell (including scode jobs):

# ~/.bashrc
module load python/anaconda-2023.09

# Optional extras your notebooks might need:
module load cuda/12.6
module load java/17.0.10 # Required by PyArrow

# Conda base env on PATH so VS Code sees *all* `conda env list` interpreters
source activate base

2.2  Per‑job Tweaks with --setup-command

If you need occasional, per‑session tweaks without touching your ~/.bashrc, pass a setup command directly to scode:

scode serve-web \
  --setup-command "module load python/anaconda-2023.09 && source activate base" \
  -- --account <pi-account> --time 01:00:00 --mem 16G

This runs just before the VS Code Server starts, ensuring both the server and any kernels it spawns see the right environment.

Why so early?

Jupyter kernels are spawned as subprocesses of the VS Code server. If you load modules after the server starts, kernels won’t inherit them. We have to restart the server (the whole job) to pick up changes.

3  Choosing An Interpreter (and How VS Code Finds Them)

According to Microsoft’s search order, the Python extension probes these locations on the server host:

Search location Typical Midway scenarios Make sure…
Virtual envs in the workspace or $HOME
(created via python -m venv venv, pipenv, etc.)		 Project‑specific venv under your repo
Venvs created in /scratch and symlinked under $HOME		 The folder contains bin/python and lives inside the workspace or $HOME
Conda envs from conda env list Anaconda module on Midway (python/anaconda-*) You loaded the module and activated base, so conda is on PATH
Sub‑folders of python.venvPath Central directory of many virtual envs Set "python.venvPath": "/path/to/venvs" in Settings (JSON)

3.1  Plain .py Files

VS Code Command Palette (Ctrl+Shift+P) → Python: Select Interpreter → pick any interpreter the VS Code Python extension found (or browse to one).

You can also click and open the interpreter selector in the far-right corner of the bottom status bar when a .py file is opened.

See Microsoft VS Code Python documentation for detailed instructions on selecting Python interpreters for .py files.

3.2  Jupyter Notebooks

The kernel‑picker lists only interpreters VS Code already knows about. You can’t type an arbitrary path. Ensure the desired venv/Conda env is discoverable according to the table above before opening the notebook, then choose it in the top‑right Select Kernel dropdown.

4  Refreshing Environments & Kernels

4.1  Running .py in Integrated Terminals

Changes to your shell environment (e.g., your ~/.bashrc) are picked up by newly reopened integrated terminals immediately. There's no need to restart the VS Code Server.

Change you made What to do...
Edited ~/.bashrc or added module load … lines Run source ~/.bashrc, or reopen a terminal tab to get the updated environment
Created a new venv/Conda env inside the running job Activate it in the terminal (e.g. source venv/bin/activate)
Installed new Python packages into the active env Ready to use immediately

4.2  Jupyter Notebooks

Notebook kernels are spawned by the VS Code Server process and inherit its environment at startup. To pick up environment changes such as an updated ~/.bashrc, you must either refresh VS Code’s environment list or restart the entire server job.

Change you made What to do...
Edited ~/.bashrc or added module load … lines Terminate the VS Code Server job (scancel …) and relaunch
Created a new venv/Conda env inside the running job Python: Refresh Environment List or reload the VS Code window
Installed new Python packages into the active env Restart the notebook kernel

5  Troubleshooting

  1. Why is my Python version not listed?

    • Run which python in an integrated terminal: is it the one you expect?
    • Check that the venv contains a bin/python and isn’t empty.
    • Verify conda env list shows your env (if using Conda).

  2. Notebook kernel stuck on “Starting”?

    • The env might miss ipykernel; run python -m pip install ipykernel.
    • Double‑check that the env’s python is executable on the compute node.

  3. Modules (e.g., CUDA) missing inside notebook but present in terminal?

    • You probably loaded them after launching the server. Pre‑load in ~/.bashrc and restart the job.

  4. Still lost?