3 minTroubleshooting

JupyterLab Issues

DAIRX provisions JupyterLab via SSH after instance boot. Most issues resolve within 3 minutes or by checking the steps below.

JupyterLab URL not appearing in dashboard

DAIRX sets up Jupyter by SSHing into your instance after it reaches "running" status. This takes 1–3 minutes depending on provider.

Wait 3 minutes after instance shows "running" — then refresh the dashboard

Check instance logs for "Jupyter setup complete" message

If instance is behind a firewall, port 8888 must be accessible

If the URL never appears after 5 minutes — the SSH provisioning may have failed. See "Manual Jupyter setup" below.

URL loads but page is blank or errors

Usually a port or token issue. The token changes every deploy.

Copy the full URL from dashboard — including the ?token= param

Try an incognito window (cached old tokens cause 403s)

If behind VPN/proxy, ensure the instance IP is reachable on port 8888

Kernel dies / GPU not detected in notebook

The Jupyter kernel runs inside the instance. If it can't find the GPU, usually a driver or CUDA mismatch.

SSH in and check:

Terminal

# Check GPU visibility
nvidia-smi

# Check CUDA version
nvcc --version

# Check PyTorch GPU access
python -c "import torch; print(torch.cuda.is_available())"

If nvidia-smi fails — NVIDIA drivers didn't load. Redeploy with a fresh instance.

If torch.cuda.is_available() is False — CUDA/PyTorch version mismatch. Use Smart Defaults to auto-match.

Manual Jupyter setup (fallback)

If automatic setup fails, SSH in and start Jupyter manually:

Terminal

# SSH into instance
ssh root@<instance_ip>

# Install (if not present)
pip install jupyterlab

# Start with remote access enabled
jupyter lab --ip 0.0.0.0 --port 8888 --no-browser --allow-root --ServerApp.token='your_secure_token'

# Access at http://<instance_ip>:8888/lab?token=your_secure_token

⚠️ Security

Use a strong token. Never use --ServerApp.token='' on a public IP.

Notebooks not saved / lost after termination

Save notebooks to /workspace — this is the default working directory

Enable Checkpoint Persistence in deploy settings — managed auto-save to Cloudflare R2 (coming soon)

Without checkpoints, terminating the instance permanently deletes all data

Back to Docs