ACTIVATE Sessions

Interactive session workflow templates for the Parallel Works ACTIVATE platform. Each workflow launches a service (Jupyter, desktop, web app, etc.), waits for it to be ready, and exposes it through the platform session proxy.

Quick Start

1. Copy the template

   cp -r workflows/hello-world workflows/my-service

2. Edit the workflow files

   • workflow.yaml - Workflow definition (inputs, jobs, session config)
   • setup.sh - Controller setup script (downloads, dependencies)
   • start.sh - Service startup script (runs on compute node)

3. Run through the ACTIVATE platform

For platform usage details, see:

• Sessions Documentation
• Building Workflows

Repository Layout

activate-sessions/
├── workflows/
│   └── hello-world/           # Template workflow (start here)
│       ├── workflow.yaml      # Workflow definition
│       ├── setup.sh           # Controller setup script
│       ├── start.sh           # Compute node service script
│       └── README.md          # Workflow documentation
├── utils/
│   └── wait_service.sh        # Shared coordination script
├── tests/                     # Pytest validation tests
├── README.md                  # This file
└── DEVELOPER_GUIDE.md         # Detailed development guide

How It Works

The Two-Script Pattern

Workflows use two scripts that run at different stages:

Script	Runs On	Purpose
setup.sh	Controller/login node	Download dependencies, pull containers, prepare shared resources
start.sh	Compute node	Start the service, write coordination files, keep session alive

Why two scripts? Controller nodes typically have internet access while compute nodes often don't. Downloads happen once on the controller before the job is submitted.

Workflow Jobs

Each workflow follows this 5-job pattern:

1. preprocessing - Checkout scripts from git, run setup.sh
2. session_runner - Submit start.sh to compute node via job_runner
3. wait_for_service - Waits for service to be ready (checks for coordination files)
4. update_session - Configures the session proxy
5. complete - Displays connection URLs

Coordination Files

Your start.sh script creates these files:

• HOSTNAME - The hostname where the service is running
• SESSION_PORT - The port the service is listening on
• job.started - Signals that the service is ready

See hello-world for a complete working example.

Running Tests

pytest -v

Documentation

• DEVELOPER_GUIDE.md - Detailed guide for creating workflows
• workflows/README.md - Workflow directory overview
• tests/README.md - Test documentation

License

MIT