TAU SLURM HPC Cluster

Maintain a VPN connection to TAU's SLURM-managed HPC cluster for submitting GPU batch jobs and monitoring their status.

Scripts location: $HOME/dotfiles/.skills/tau-hpc/scripts/

Property	Value
Host	`slurmlogin.tau.ac.il`
Username	`galaharoni`
Account	`gpu-tad-users_v2`
Working directory	`/scratch300/galaharoni`
Docs	https://hpcguide.tau.ac.il/

Important: The default home directory (/a/home/cc/students/cs/galaharoni) doesn't exist. Always use /scratch300/galaharoni and set HOME accordingly.

Quick Start

1. Connect VPN

bash

# Set credentials
export TAU_USERNAME=galaharoni
export TAU_PASSWORD=<password>

# Get OTP from user, then connect (uv handles dependencies automatically)
uv run $HOME/dotfiles/.skills/tau-hpc/scripts/vpn-connect.py $TAU_USERNAME $TAU_PASSWORD <OTP>

# Verify connection
ip addr show tun0  # Should show 10.x.x.x address
pgrep openconnect  # Should show running process

2. SSH to Cluster

bash

# Run command on cluster (sets HOME correctly)
sshpass -p "$TAU_PASSWORD" ssh -o StrictHostKeyChecking=no galaharoni@slurmlogin.tau.ac.il "
export HOME=/scratch300/galaharoni
cd \$HOME
<your commands here>
"

3. Submit GPU Job

bash

sshpass -p "$TAU_PASSWORD" ssh -o StrictHostKeyChecking=no galaharoni@slurmlogin.tau.ac.il "
export HOME=/scratch300/galaharoni
cd /scratch300/galaharoni/your-repo

cat > job.sbatch << 'EOF'
#!/bin/bash
#SBATCH --job-name=my-job
#SBATCH --partition=gpu-general-pool
#SBATCH --qos=public
#SBATCH --gres=gpu:1
#SBATCH --cpus-per-task=4
#SBATCH --mem=32G
#SBATCH --time=01:00:00
#SBATCH --output=/scratch300/galaharoni/job_%j.out

export HOME=/scratch300/galaharoni
export PATH=\$HOME/.local/bin:\$PATH
cd /scratch300/galaharoni/your-repo

uv sync
uv run python -u your_script.py
EOF

sbatch job.sbatch
squeue -u galaharoni
"

4. Check Job Output

bash

sshpass -p "$TAU_PASSWORD" ssh -o StrictHostKeyChecking=no galaharoni@slurmlogin.tau.ac.il "
tail -50 /scratch300/galaharoni/job_<JOBID>.out
"

VPN Connection

TAU uses Palo Alto GlobalProtect VPN with SAML authentication.

Requirements

bash

# System packages
sudo apt install openconnect sshpass

# Install gpclient 2.5.x from https://github.com/yuezk/GlobalProtect-openconnect/releases
sudo apt install ./globalprotect-openconnect_*.deb

# First-time only: install playwright browser (uv handles the Python package)
uv run --with playwright python -c "import playwright; playwright.sync_api.sync_playwright().start().chromium.launch()"

How vpn-connect.py Works

•Starts gpclient in gateway mode with --browser remote
•gpclient creates a local auth server (e.g., http://10.x.x.x:PORT/UUID)
•Playwright opens the auth URL which redirects to TAU's SAML IdP
•Fills credentials and OTP automatically
•Extracts callback URL from the "click here" link on the auth complete page
•Decodes the cookie from the base64-encoded callback data
•Connects via openconnect with --background flag (keeps VPN alive)

Usage

bash

# With all arguments
uv run $HOME/dotfiles/.skills/tau-hpc/scripts/vpn-connect.py galaharoni <password> <OTP>

# With environment variables
export TAU_USERNAME=galaharoni
export TAU_PASSWORD=<password>
uv run $HOME/dotfiles/.skills/tau-hpc/scripts/vpn-connect.py <OTP>

Verify Connection

bash

ip addr show tun0              # Check tunnel interface
pgrep -a openconnect           # Check process
sshpass -p "$TAU_PASSWORD" ssh galaharoni@slurmlogin.tau.ac.il hostname  # Test SSH

Disconnect

bash

sudo pkill openconnect

Troubleshooting

Issue	Solution
"No auth URL"	gpclient may have crashed; try again
"No callback"	OTP expired (they're short-lived); get fresh OTP
"No tun0 interface"	openconnect failed; check sudo permissions
SSH timeout	VPN not connected; re-run vpn-connect.py

SSH Setup

SSH keys don't work because the home directory doesn't exist. Use sshpass.

Non-Interactive Commands

Always set HOME explicitly:

bash

sshpass -p "$TAU_PASSWORD" ssh -o StrictHostKeyChecking=no galaharoni@slurmlogin.tau.ac.il "
export HOME=/scratch300/galaharoni
export PATH=\$HOME/.local/bin:\$PATH
cd \$HOME
<commands>
"

Copying Files to Cluster

rsync is not available. Use scp with tar:

bash

# Create tarball (exclude large directories)
cd /path/to/repo
tar czf /tmp/repo.tar.gz --exclude='.venv' --exclude='__pycache__' --exclude='.git' --exclude='results' .

# Copy to cluster
sshpass -p "$TAU_PASSWORD" scp -o StrictHostKeyChecking=no /tmp/repo.tar.gz galaharoni@slurmlogin.tau.ac.il:/scratch300/galaharoni/

# Extract on cluster
sshpass -p "$TAU_PASSWORD" ssh -o StrictHostKeyChecking=no galaharoni@slurmlogin.tau.ac.il "
export HOME=/scratch300/galaharoni
cd /scratch300/galaharoni
mkdir -p my-repo && cd my-repo
tar xzf ../repo.tar.gz
"

GPU Resources

Partition	QOS	GPUs	Notes
`gpu-tad-pool`	`owner`	8x H100, 32x A100	Priority access
`gpu-general-pool`	`public`	~117 GPUs (H100, A100, L40S, RTX A6000)	More availability

Check Available Resources

bash

sinfo -p gpu-general-pool -o "%P %a %D %t %G"

Job Management

bash

squeue -u galaharoni                              # Check queue
scancel <JOBID>                                   # Cancel job
sacct -j <JOBID>                                  # Job history
tail -f /scratch300/galaharoni/job_<JOBID>.out    # Watch output

Tools Available on Cluster

•uv: /scratch300/galaharoni/.local/bin/uv (Python package manager)
•gh: GitHub CLI (in .local/bin)