AI Power Progress iA
Cluster onboarding

Windows Worker (GPU node)

Bring your Windows node online for Ray + Ollama so the cluster sees WINDOWS_WORKER_HOST and the website stops reporting the GPU worker as missing.

Checking status...

Step 1: Serve the bootstrap scripts (head)

On the head node (HEAD_NODE_HOST), run:

cd $HOME
./serve_windows_bootstrap.sh start 8088 7200
./serve_windows_bootstrap.sh status

This serves bootstrap_windows_worker.ps1 and install_ollama.ps1 over HTTP to make onboarding copy/paste simple.

Step 2A: Bootstrap the Windows node (PowerShell as Administrator)

Run this on the Windows machine (WINDOWS_WORKER_HOST):

Set-ExecutionPolicy Bypass -Scope Process -Force
$head = 'HEAD_NODE_HOST'
$port = 8088
$out  = Join-Path $env:TEMP 'bootstrap_windows_worker.ps1'
Invoke-WebRequest "http://${head}:$port/bootstrap_windows_worker.ps1" -OutFile $out -UseBasicParsing
powershell.exe -NoProfile -ExecutionPolicy Bypass -File $out -HeadIp $head -HeadHttpPort $port -RayPort 6380 -EnableSsh -InstallHeadSshKey -SshAuthorizedKeysPath "$env:USERPROFILE\.ssh\authorized_keys" -ConfigureFirewall -PurgeRayWorkerVenv -InstallPython

This enables OpenSSH (key-based), installs Python 3.12 + Ray, opens required ports, and installs a persistent Ray worker scheduled task.

Step 2B (optional): Bootstrap over SSH from the head (prompts for password once)

If Windows already has OpenSSH Server enabled and you can log in with a password, you can run the bootstrap remotely from the head:

cd $HOME
./bootstrap_windows_remote.sh WINDOWS_WORKER_HOST
  • This prompts for your Windows SSH password in your terminal and does not store it.
  • After bootstrap, key-based SSH should work (BatchMode=yes), so automation can manage the worker.

Step 3: Install/repair Ollama (LAN-serving) on Windows (recommended)

Set-ExecutionPolicy Bypass -Scope Process -Force
$head = 'HEAD_NODE_HOST'
$port = 8088
$out  = Join-Path $env:TEMP 'install_ollama.ps1'
Invoke-WebRequest "http://${head}:$port/install_ollama.ps1" -OutFile $out -UseBasicParsing
powershell.exe -NoProfile -ExecutionPolicy Bypass -File $out -BindHost LAN_HOST -ConfigureFirewall -AllowFrom $head -PullModels

After this, the head should be able to reach http://WINDOWS_WORKER_HOST:11434/api/version.

Step 4: Validate from the head

cd $HOME
./cluster_excellence_run.sh --repair
python3 ray_test.py

Notes

  • This workflow does not write to your USB boot drive.
  • If SSH still prompts for password, re-run Step 2 and confirm OpenSSH Server is enabled and the head key was installed into authorized_keys.
  • If remote Ollama stays down, run Step 3 again and confirm Windows Firewall allows TCP 11434 from the head.