Skip to main content

VS Code + Container Setup Guide

Overview​

There are two ways to use VS Code with DS01 containers. Choose based on your workflow:

  1. Remote SSH (Edit on host, run in container via terminal)
  2. Dev Containers (Entire VS Code session runs inside container) ← RECOMMENDED

Method 1: Remote SSH (Traditional)​

What happens:

  • VS Code connects to DS01 host via SSH
  • You edit files in ~/workspace/your-project
  • You manually enter containers via integrated terminal
  • Code runs in container, but IDE is on host

Setup:

# 1. In VS Code, install "Remote - SSH" extension

# 2. Connect to DS01
# Cmd+Shift+P → "Remote-SSH: Connect to Host"
# Enter: username@ds01-server

# 3. Open your project folder
# File → Open Folder → ~/workspace/your-project

# 4. To run code, open terminal and enter container:
container-run my-project

Pros:

  • Simple setup
  • Edit multiple projects simultaneously
  • Easy to switch between containers

Cons:

  • Python intellisense uses HOST Python (not container)
  • Debugger runs on HOST (not in container)
  • Jupyter kernels are on HOST (not container GPUs)
  • NOT A TRUE CONTAINERIZED WORKFLOW

What happens:

  • VS Code connects DIRECTLY to running container
  • Entire IDE session runs inside container
  • Python interpreter, debugger, Jupyter ALL use container
  • GPU-accelerated code execution
  • TRUE CONTAINERIZED WORKFLOW ✓

Setup Steps:​

1. Install Extensions (on your local machine)​

- Dev Containers (ms-vscode-remote.remote-containers)
- Remote - SSH (ms-vscode-remote.remote-ssh)

2. Start Your Container on DS01​

# SSH to DS01 first
ssh username@ds01-server

# Create/start your container
container-run my-project
# Exit with Ctrl+P, Ctrl+Q (keeps it running)

3. Connect VS Code to Container​

Option A: Via SSH Tunnel (Recommended)

# On your local machine, create SSH tunnel:
ssh -L 2222:localhost:2222 username@ds01-server

# In another terminal on DS01:
docker exec -it my-project._.$(id -u) socat TCP-LISTEN:2222,fork,reuseaddr UNIX-CONNECT:/var/run/docker.sock

# In VS Code:
# Cmd+Shift+P → "Dev Containers: Attach to Running Container..."
# Select your container

Option B: Direct Attach (if on DS01 network)

1. Cmd+Shift+P → "Remote-SSH: Connect to Host" → ds01-server
2. Cmd+Shift+P → "Dev Containers: Attach to Running Container..."
3. Select container: my-project._.1001

4. Verify Container Session​

Once connected, verify you're INSIDE the container:

# In VS Code integrated terminal:
pwd
# Should show: /workspace

hostname
# Should show your container name

which python
# Should show container's Python, not host

nvidia-smi
# Should show GPU (if container has GPU access)

Working in Dev Container Mode:​

File Editing:

  • Files in /workspace are your persistent workspace
  • Edit directly in VS Code

Running Code:

  • Terminal → runs inside container automatically
  • Python interpreter → uses container's Python
  • Debugger → debugs inside container
  • Jupyter → kernels run in container with GPU access

Installing Packages:

# In VS Code terminal (already inside container):
pip install transformers
# Installs to container environment

GPU Access:

import torch
print(torch.cuda.is_available())  # Should be True

Verifying Your Setup​

✓ Correct Setup (Running in Container):​

# Check prompt
user@my-project:/workspace$  # Container hostname

# Check Python
which python
# /opt/conda/bin/python (or similar container path)

# Check packages
pip list | grep torch
# Shows container's packages

# Check workspace
ls /workspace
# Shows your project files

✗ Wrong Setup (Running on Host):​

# Check prompt
user@ds01:/home/username/workspace$  # DS01 hostname

# Check Python
which python
# /usr/bin/python or /home/username/.local/...

# No GPU access in Python
python -c "import torch; print(torch.cuda.is_available())"
# False or module not found

For ML Development:​

  1. Use Dev Containers (Method 2)
  2. Connect VS Code directly to running container
  3. All development happens in container
  4. GPU acceleration works automatically
  5. Dependencies are isolated

For Multi-Project Editing:​

  1. Use Remote SSH (Method 1)
  2. Edit files on host
  3. Run code by entering containers as needed
  4. Good for quick edits across projects

Common Issues​

"Cannot find Python interpreter"​

  • Cause: VS Code is on host, not in container
  • Fix: Use Dev Containers to attach to container directly

"CUDA not available"​

  • Cause: Python running on host, not container
  • Fix: Verify you're in container (check hostname)

"Package not found"​

  • Cause: Using host Python, not container Python
  • Fix: Use Dev Containers or manually enter container first

"Permission denied accessing /var/run/docker.sock"​

  • Cause: Container doesn't have docker socket mounted
  • Fix: Recreate container with container-create (new containers have it)

"Command not found inside container"​

  • Cause: Old container created before DS01 updates
  • Fix: Create new container - all host commands now work inside containers too

Summary​

For true containerized ML workflows: ✓ Use Dev Containers extension ✓ Attach to running container ✓ Verify with hostname and which python ✓ All code runs in container with GPU access ✓ All DS01 commands work inside containers (container-list, image-list, etc.)

Quick editing across projects: ✓ Use Remote SSH ✓ Enter containers manually when running code ✓ Can still manage containers from inside with container-list, container-run

Note: As of latest updates, ALL host commands are available inside containers through docker socket mounting. This means you can manage containers from within containers, list images, etc. - providing a consistent command experience regardless of context.

Choose based on your needs!