release-process.md

Release Process

This document describes how to create and publish a release of Fast Model Actuation (FMA).

Release Types

Regular Release

For stable, production-ready versions:

• Tag format: v0.3.1, v1.0.0, etc.
• Mark as "Latest release" in GitHub

Pre-Release (Test Release)

For testing before official release:

• Tag format: v0.3.0-alpha.1, v0.3.0-rc.1, v0.3.0-beta.2, etc.
• Mark as "Pre-release" in GitHub
• Useful for integration testing in llm-d-benchmark or other environments

Step-by-Step Release Process

1. Create and Push the Git Tag

Ensure all changes for the release are merged to main. Then, create and push the Git tag:

git tag v0.3.1 # regular release (recommended format)

git tag v0.3.0-alpha.1 # pre-release

git push origin v0.3.1  # or your tag name

2. Create the GitHub Release

1. Go to https://github.com/llm-d-incubation/llm-d-fast-model-actuation/releases/new

2. Choose a tag: Select the tag you just pushed (e.g., v0.3.1)

3. Release title: Use the tag name (e.g., v0.3.1)

4. Description: Document what's new in this release:

   ## What's Changed
   - Feature: Added support for X
   - Fix: Resolved issue with Y
   - Improvement: Enhanced Z performance
   
   ## Breaking Changes
   - Changed API for ...
   
   ## Upgrade Notes
   - Users should ...

5. Pre-release checkbox:

   • ✅ Check for pre-releases (alpha, beta, rc)
   • ⬜ Leave unchecked for regular releases

6. Click "Publish release"

3. Automated Workflow Execution

Once you publish the release, the publish-release workflow runs automatically and performs the following steps:

1. Builds 4 container images:

   • ghcr.io/llm-d-incubation/llm-d-fast-model-actuation/dual-pods-controller:v0.3.1
   • ghcr.io/llm-d-incubation/llm-d-fast-model-actuation/launcher-populator:v0.3.1
   • ghcr.io/llm-d-incubation/llm-d-fast-model-actuation/launcher:v0.3.1
   • ghcr.io/llm-d-incubation/llm-d-fast-model-actuation/requester:v0.3.1

2. Updates Helm chart values with the release-specific image references

3. Packages Helm charts using helm package --version 0.3.1 --app-version v0.3.1

4. Publishes chart to GHCR:

   • oci://ghcr.io/llm-d-incubation/llm-d-fast-model-actuation/charts/fma-controllers

4. Manual Testing

For testing purposes, the workflow can be manually triggered:

Use the GitHub Actions UI to manually trigger the workflow with a specific tag

1. Go to Actions → publish release → Run workflow
2. Enter the tag (e.g., v0.3.1)
3. Click "Run workflow"

Related Documentation

• Workflow Source - The actual workflow implementation