VSCode Tunneling on Pantarhei

VSCode Tunneling on Pantarhei

VS Code Remote Tunnels let you connect your local VS Code instance directly to a Pantarhei compute node. Once the tunnel is running, you get a full IDE experience—file editing, terminal access, extensions, and debugger—over a secure connection, without needing an open inbound port or a VPN.

Compute nodes have direct internet access, so the tunnel can reach Microsoft's relay servers without a proxy.

Prerequisites

• A Pantarhei account with access to the normal partition
• Visual Studio Code installed on your local machine
• The Remote - Tunnels extension installed in your local VS Code (ms-vscode.remote-server)
• A GitHub or Microsoft account (used to authenticate the tunnel on first use)

Job Script

Save the following script as vscode.job in your Pantarhei home directory:

#!/bin/bash
# vscode.job
#SBATCH --job-name=VSCode
#SBATCH --output=VSCode-%J.txt
#SBATCH --nodes=1
#SBATCH --ntasks=32
#SBATCH --partition=normal
#SBATCH --time=00:30:00
#SBATCH --mail-type=ALL
#SBATCH --mail-user=your_email@example.com

# Compute nodes have direct internet access.
# Unset proxy variables in case they are inherited from the login node.
unset HTTP_PROXY HTTPS_PROXY http_proxy https_proxy

export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK

echo "Cluster:      $SLURM_CLUSTER_NAME"
echo "Tasks/node:   $SLURM_TASKS_PER_NODE"
echo "CPUs/task:    $SLURM_CPUS_PER_TASK"
echo "CPUs on node: $SLURM_CPUS_ON_NODE"
echo "Total tasks:  $SLURM_NTASKS"
echo "Nodes:        $SLURM_NODELIST"
echo "Num nodes:    $SLURM_NNODES"
echo "GPUs:         $CUDA_VISIBLE_DEVICES"

module load VSCode

code tunnel --verbose --accept-server-license-terms

Replace your_email@example.com with your actual email address so Slurm can send you job notifications.

Script Explanation

SBATCH directives

Directive	Value	Purpose
--job-name=VSCode	VSCode	Name shown in squeue output
--output=VSCode-%J.txt	VSCode-<jobid>.txt	Captures all stdout/stderr; %J is replaced with the job ID at runtime
--nodes=1	1	Requests a single compute node — a tunnel only needs one
--ntasks=32	32	Number of CPU tasks available to your session
--partition=normal	normal	Submits to the standard CPU queue
--time=00:30:00	30 min	Wall-time limit; the tunnel stops when this expires
--mail-type=ALL	ALL	Sends email when the job starts, ends, or fails
--mail-user	your email	Address that receives the Slurm notifications

Proxy variables

unset HTTP_PROXY HTTPS_PROXY http_proxy https_proxy

Login nodes may set proxy environment variables for outbound HTTP traffic. Compute nodes have direct internet access, so those proxies are not needed and can actually block the tunnel from reaching Microsoft's relay servers. This block removes them explicitly.

OpenMP thread count

export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK

Sets the number of OpenMP threads to match the CPUs allocated per task, preventing accidental over-subscription if you run multithreaded code inside the VS Code terminal.

Diagnostic echo statements

echo "Cluster:      $SLURM_CLUSTER_NAME"
echo "Tasks/node:   $SLURM_TASKS_PER_NODE"
...

These lines print key Slurm environment variables to the output file (VSCode-<jobid>.txt). They are useful for confirming which node and resources were actually allocated to your job.

Module load

module load VSCode

Loads the VS Code CLI (code) into PATH using Pantarhei's module system. Without this, the code command would not be found on the compute node.

Tunnel command

code tunnel --verbose --accept-server-license-terms

Flag	Purpose
tunnel	Starts a VS Code Remote Tunnel that relays traffic through Microsoft's servers
--verbose	Writes detailed status messages to the output file, including the authentication URL and tunnel name
--accept-server-license-terms	Non-interactively accepts the VS Code Server license so the job does not stall waiting for keyboard input

Submitting the Job

Submit the job from the Pantarhei login node:

sbatch vscode.job

Slurm will return a job ID, for example:

Submitted batch job 312

First-Time Authentication

The first time code tunnel runs it needs to authenticate against a Microsoft or GitHub account. The authentication link is written to the job's output file.

Check the output file (replace 312 with your job ID):

cat VSCode-312.txt

Look for a line like:

To grant access to the server, please log into https://github.com/login/device and use code XXXX-XXXX

Open that URL in your browser, enter the device code, and complete the sign-in. The tunnel will start automatically once authentication succeeds. Subsequent job runs on the same account do not require this step.

Connecting from Your Local Machine

After authentication, the output file will show the tunnel name and connection URL, for example:

Open this link in your browser https://vscode.dev/tunnel/<tunnel-name>/

Option 1 — Browser: Open the URL directly in a browser for a web-based VS Code session.

Option 2 — Local VS Code:

1. Open VS Code on your local machine.
2. Press Ctrl+Shift+P (or Cmd+Shift+P on macOS) to open the Command Palette.
3. Run Remote Tunnels: Connect to Tunnel. If the Remote - Tunnels extension is not installed, install it first from the Extensions marketplace.
4. Sign in with the same GitHub or Microsoft account used on the cluster.
5. Select the tunnel named after your Pantarhei session.

You now have a full VS Code window connected to the compute node. The integrated terminal runs directly on the node, so you can run scripts, inspect files, and use all installed modules.

Monitoring the Job

Check whether the tunnel job is still running:

squeue -u $USER

View live output from the job:

tail -f VSCode-<JOBID>.txt

Ending the Session

The tunnel stops when the Slurm job ends (after the 4-hour wall time) or when you cancel it manually:

scancel <JOBID>

You can also stop the tunnel from within VS Code by running Remote Tunnels: Close Remote Connection from the Command Palette.

Adjusting Resources

The default script requests 32 tasks for up to 4 hours on the normal partition. Adjust the #SBATCH directives to match your workload:

Directive	Default	Notes
--ntasks	32	Reduce if you only need a light editing session
--time	00:30:00	Maximum on normal is 7 days
--partition	normal	Use gpu or gpu-more-time for GPU work
--mem	(not set)	Add --mem=64G if your work is memory-intensive

Troubleshooting

Tunnel URL does not appear in the output file

The job may still be starting. Wait a moment and re-run tail -f VSCode-<JOBID>.txt. If nothing appears after a few minutes, check that the VSCode module loaded successfully and that no proxy variables interfered.

Authentication prompt does not appear

If you have authenticated before, the tunnel skips the prompt and starts immediately. Look for the Open this link line instead of the device-code line.

Connection drops after a short time

The tunnel stays alive as long as the Slurm job is running. If the job hits its wall-time limit (--time), the tunnel will close. Increase --time or resubmit the job.

module load VSCode fails

Run module avail VSCode on the login node to confirm the module name. Contact your system administrator if the module is not listed.

Additional Resources

For complete example scripts and job files, visit the Pantarhei examples repository:

https://github.com/kamalcou/pantarhei-examples