Skip to content

Latest commit

 

History

History
271 lines (188 loc) · 9.91 KB

VERSIONING.md

File metadata and controls

271 lines (188 loc) · 9.91 KB

Versioning and Branching in controller-runtime

NB: this also applies to controller-tools.

TL;DR:

Users

  • We follow Semantic Versioning (semver)
  • Use releases with your dependency management to ensure that you get compatible code
  • The master branch contains all the latest code, some of which may break compatibility (so "normal" go get is not recommended)

Contributors

  • All code PR must be labeled with 🐛 (patch fixes), ✨ (backwards-compatible features), or ⚠️ (breaking changes)

  • Breaking changes will find their way into the next major release, other changes will go into an semi-immediate patch or minor release

  • Please try to avoid breaking changes when you can. They make users face difficult decisions ("when do I go through the pain of upgrading?"), and make life hard for maintainers and contributors (dealing with differences on stable branches).

Mantainers

Don't be lazy, read the rest of this doc :-)

Overview

controller-runtime (and friends) follow Semantic Versioning. I'd recommend reading the aforementioned link if you're not familiar, but essentially, for any given release X.Y.Z:

  • an X (major) release indicates a set of backwards-compatible code. Changing X means there's a breaking change.

  • a Y (minor) release indicates a minimum feature set. Changing Y means the addition of a backwards-compatible feature.

  • a Z (patch) release indicates minimum set of bugfixes. Changing Z means a backwards-compatible change that doesn't add functionality.

NB: If the major release is 0, any minor release may contain breaking changes.

These guarantees extend to all code exposed in public APIs of controller-runtime. This includes code both in controller-runtime itself, plus types from dependencies in public APIs. Types and functions not in public APIs are not considered part of the guarantee.

In order to easily maintain the guarantees, we have a couple of processes that we follow.

Branches

controller-runtime contains two types of branches: the master branch and release-X branches.

The master branch is where development happens. All the latest and greatest code, including breaking changes, happens on master.

The release-X branches contain stable, backwards compatible code. Every major (X) release, a new such branch is created. It is from these branches that minor and patch releases are tagged. If some cases, it may be necessary open PRs for bugfixes directly against stable branches, but this should generally not be the case.

The maintainers are responsible for updating the contents of this branch; generally, this is done just before a release using release tooling that filters and checks for changes tagged as breaking (see below).

Tooling

  • release-notes.sh: generate release notes for a range of commits, and check for next version type (TODO)

  • verify-commit-messages.sh: check that your PR and/or commit messages have the right versioning icon (TODO).

PR Process

Every PR should be annotated with an icon indicating whether it's a:

  • Breaking change: ⚠️ (:warning:)
  • Non-breaking feature: ✨ (:sparkles:)
  • Patch fix: 🐛 (:bug:)
  • Docs: 📖 (:book:)
  • Infra/Tests/Other: 🏃 (:running:)
  • No release note: 👻 (:ghost:)

Use 👻 (no release note) only for the PRs that change or revert unreleased changes, which don't deserve a release note. Please don't abuse it.

You can also use the equivalent emoji directly, since GitHub doesn't render the :xyz: aliases in PR titles.

Individual commits should not be tagged separately, but will generally be assumed to match the PR. For instance, if you have a bugfix in with a breaking change, it's generally encouraged to submit the bugfix separately, but if you must put them in one PR, mark the commit separately.

Commands and Workflow

controller-runtime follows the standard Kubernetes workflow: any PR needs lgtm and approved labels, PRs authors must have signed the CNCF CLA, and PRs must pass the tests before being merged. See the contributor docs for more info.

We use the same priority and kind labels as Kubernetes. See the labels tab in GitHub for the full list.

The standard Kubernetes comment commands should work in controller-runtime. See Prow for a command reference.

Release Process

Minor and patch releases are generally done immediately after a feature or bugfix is landed, or sometimes a series of features tied together.

Minor releases will only be tagged on the most recent major release branch, except in exceptional circumstances. Patches will be backported to maintained stable versions, as needed.

Major releases are done shortly after a breaking change is merged -- once a breaking change is merged, the next release must be a major revision. We don't intend to have a lot of these, so we may put off merging breaking PRs until a later date.

Exact Steps

Follow the release-specific steps below, then follow the general steps after that.

Minor and patch releases

  1. Update the release-X branch with the latest set of changes by calling git rebase master from the release branch.

Major releases

  1. Create a new release branch named release-X (where X is the new version) off of master.

General

  1. Generate release notes using the release note tooling.

  2. Add a release for controller-runtime on GitHub, using those release notes, with a title of vX.Y.Z.

  3. Do a similar process for controller-tools

  4. Announce the release in #kubebuilder on Slack with a pinned message.

  5. Potentially update kubebuilder as well.

Breaking Changes

Try to avoid breaking changes. They make life difficult for users, who have to rewrite their code when they eventually upgrade, and for maintainers/contributors, who have to deal with differences between master and stable branches.

That being said, we'll occasionally want to make breaking changes. They'll be merged onto master, and will then trigger a major release (see Release Process). Because breaking changes induce a major revision, the maintainers may delay a particular breaking change until a later date when they are ready to make a major revision with a few breaking changes.

If you're going to make a breaking change, please make sure to explain in detail why it's helpful. Is it necessary to cleanly resolve an issue? Does it improve API ergonomics?

Maintainers should treat breaking changes with caution, and evaluate potential non-breaking solutions (see below).

Note that API breakage in public APIs due to dependencies will trigger a major revision, so you may occasionally need to have a major release anyway, due to changes in libraries like k8s.io/client-go or k8s.io/apimachinery.

NB: Pre-1.0 releases treat breaking changes a bit more lightly. We'll still consider carefully, but the pre-1.0 timeframe is useful for converging on a ergonomic API.

Avoiding breaking changes

Solutions to avoid
  • Confusingly duplicate methods, functions, or variables.

    For instance, suppose we have an interface method List(ctx context.Context, options *ListOptions, obj runtime.Object) error, and we decide to switch it so that options come at the end, parametrically. Adding a new interface method ListParametric(ctx context.Context, obj runtime.Object, options... ListOption) is probably not the right solution:

    • Users will intuitively see List, and use that in new projects, even if it's marked as deprecated.

    • Users who don't notice the deprecation may be confused as to the difference between List and ListParametric.

    • It's not immediately obvious in isolation (e.g. in surrounding code) why the method is called ListParametric, and may cause confusion when reading code that makes use of that method.

    In this case, it may be better to make the breaking change, and then eventually do a major release.

Why don't we...

Use "next"-style branches

Development branches:

  • don't win us much in terms of maintenance in the case of breaking changes (we still have to merge/manage multiple branches for development and stable)

  • can be confusing to contributors, who often expect master to have the latest changes.

Never break compatibility

Never doing a new major release could be an admirable goal, but gradually leads to API cruft.

Since one of the goals of controller-runtime is to be a friendly and intuitive API, we want to avoid too much API cruft over time, and occaisonal breaking changes in major releases help accomplish that goal.

Furthermore, our dependency on Kubernetes libraries makes this difficult (see below)

Always assume we've broken compatibility

a.k.a. k8s.io/client-go style

While this makes life easier (a bit) for maintainers, it's problematic for users. While breaking changes arrive sooner, upgrading becomes very painful.

Furthermore, we still have to maintain stable branches for bugfixes, so the maintenance burden isn't lessened by a ton.

Extend compatibility guarantees to all dependencies

This is very difficult with the number of Kubernetes dependencies we have. Kubernetes dependencies tend to either break compatibility every major release (e.g. k8s.io/client-go, which loosely follows semver), or at a whim (many other Kubernetes libraries).

If we limit to the few objects we expose, we can better inform users about how controller-runtime itself has changed in a given release. Then, users can make informed decisions about how to proceed with any direct uses of Kubernetes dependencies their controller-runtime-based application may have.