Skip to content

Latest commit

 

History

History
110 lines (76 loc) · 5.25 KB

RELEASE.md

File metadata and controls

110 lines (76 loc) · 5.25 KB
Raw

Release

Creating a release

This release process itself should be as automated as possible, and has only a few steps:

  1. Trigger a new release with make release. At this point you'll see a preview changelog in the terminal. If you're happy with the changelog, press y to continue, otherwise you can abort and adjust the labels on the PRs and issues to be included in the release and re-run the release trigger command.

  2. A release admin must approve the release on the GitHub Actions release pipeline run page. Once approved, the release pipeline will generate all assets and publish a GitHub Release.

  3. If there is a release Milestone, close it.

Ideally releasing should be done often with small increments when possible. Unless a breaking change is blocking the release, or no fixes/features have been merged, a good target release cadence is between every 1 or 2 weeks.

Retracting a release

If a release is found to be problematic, it can be retracted with the following steps:

  • Deleting the GitHub Release
  • Add a new retract entry in the go.mod for the versioned release

Note: do not delete release tags from the git repository since there may already be references to the release in the go proxy, which will cause confusion when trying to reuse the tag later (the H1 hash will not match and there will be a warning when users try to pull the new release).

Background

A good release process has the following qualities:

  1. There is a way to plan what should be in a release
  2. There is a way to see what is actually in a release
  3. Allow for different kinds of releases (major breaking vs backwards compatible enhancements vs patch updates)
  4. Specify a repeatable way to build and publish software artifacts

Planning a release

To indicate a set of features to be released together add each issue to an in-repository Milestone named with major-minor version to be released (e.g. v0.1). It is OK for other features to be in the release that were not originally planned, and these issues and PRs do not need to be added to the Milestone in question. Only the set of features that, when completed, would allow the release to be considered complete. A Milestone is only used to:

  • Plan what is desired to be in a release
  • Track progress to indicate when we may be ready to cut a new release

Not all releases need to be planned. For instance, patch releases for fixes should be released when they are ready and when releasing would not interfere with another current release (where some partial or breaking features have already been merged).

Unless necessary, feature releases should be small and frequent, which may obviate the need for regular release planning under a Milestone.

What is in a release

Milestones are specifically for planning a release, not necessarily tracking all changes that a release may bring (and more importantly, not all releases are necessarily planned either).

This is one of the (many) reasons for a Changelog. A good Changelog lists changes grouped by the type of change (new, enhancement, deprecation, breaking, bug fix, security fix), in chronological order (within groups), linking the PR where the change was made in the Changelog line. Furthermore, there should be a place to see all released versions, the release date for each release, the semantic version of the release, and the set of changes for each release.

This project auto-generates the Changelog contents for each current release and posts the generated contents to the GitHub Release page. Leveraging the GitHub Releases feature allows GitHub to manage the Changelog on each release outside of the git source tree while still being hosted with the released assets.

The Changelog is generated from the metadata from in-repository issues and PRs, using labels to guide what kind of change each item is (e.g. breaking, new feature, bug fix, etc). Only issues/PRs with select labels are included in the Changelog, and only if the issue/PR was created after the last release. Additional labels are used to exclude items from the Changelog.

The above suggestions imply that we should:

  • Ensure there is a sufficient title for each PR and issue title to be included in the Changelog
  • The appropriate label is applied to PRs and/or issues to drive specific change type sections (deprecated, breaking, security, bug, etc)

With this approach as we cultivate good organization of PRs and issues we automatically get an equally good Changelog.

Major, minor, and patch releases

The latest version of the tool is the only supported version, which implies that multiple parallel release branches will not be a regular process (if ever). Multiple releases can be planned in parallel, however, only one can be actively developed at a time. That is, if PRs attached to a release Milestone have been merged into the main branch, that release is now the "next" release. This implies that the source of truth for release lies with the git log and Changelog, not with the release Milestones (which are purely for planning and tracking).

Semantic versioning should be used to indicate breaking changes, new features, and fixes. The exception to this is < 1.0, where the major version is not bumped for breaking changes, instead the minor version indicates both new features and breaking changes.