This document is meant to explain the release process of the AsyncAPI specification. It aims to describe all details of the process so that any member of the community can jump in and help coordinate.
It covers:
- Who? - who is responsible for doing what in a release
- When? - when releases happen
- What? - what is covered by this document
- How? - a description of the steps needed for a release
AsyncAPI specification release involves not only spec release but also the release of AsyncAPI-maintained tools. Therefore it requires a single person called release coordinator to ensure the release goes well throughout all phases. The release coordinator is responsible for working through the process described below.
The release coordinator doesn't have to do all the work alone. The release coordinator needs to find the right people to do the work required for a given phase and engage as many people as needed.
Each release can have a different release coordinator. If you are interested and would like to know more, join our Slack workspace and visit the 11_how-to-contribute channel.
Many of the steps needed to release a new version of the AsyncAPI specification (e.g. merging, creating branches, creating releases) need support from administrators for individual github.com/asyncapi repositories.
Each respository contains a CODEOWNERS file (e.g. https://github.com/asyncapi/parser-js/blob/master/CODEOWNERS) that identifies the people who can help.
AsyncAPI releases happen regularly in the following months:
- January
- April
- June
- September
Regular releases concern only major and minor releases. We do not decide up front if the next release is major or minor. This decision depends on the proposals for changes in the specification and how much they affect specification and tooling in a given release cycle.
Patch releases are automatically released and published.
The release is scheduled for a given month, not a specific day. We will release once everything is ready, and won't wait until the last day of the month with the release.
We release often, so anything that isn't ready can wait until the next release (e.g. it can be released in June if the change will not make it in April). Our priority is the quality, and not quantity, of the features. Quality means that changes for a release is properly delivered in as many tools as possible with very good documentation and examples.
Github repositories that need to be updated at the same time to enable a release, in a coordinated way, are:
- spec - the specification file
- spec-json-schemas - the JSON schema
- parser-js - the JavaScript parser
- website - the website
The steps described here are not intended to be strictly chronological. For a single release, it might make sense to start some pieces of work in a slightly different order. The intention of the steps here is meant to give release coordinators a rough idea of the sequence involved.
A nice way to start a new release is to do one or more of the initial steps on an open live call/meeting.
This isn't a required step, but possible benefits include:
- raising awareness of the upcoming release
- giving a new release coordinator support and initial direction to help them get started
- allowing potential future release coordinators to see what is involved
Examples of arranging a contribution call like this can be found in the community repository
To allow the community to see what is currently being considered for inclusion in a release, and to give easy, at-a-glance visibility over the progress, create a new Github issue in the spec repository.
This issue should be updated as each of the following steps are completed.
A template for a new release issue is available.
An example is the release issue for the 2.3.0 release.
Release branches must have a year and a month of the release as prefix: {YEAR_OF_RELEASE}-{MONTH_OF_RELEASE}-release.
For example, a release created in September 2021 has a 2021-09-release release branch.
At the beginning of the release cycle, we need to have a new release branch created in the following repositories:
- spec where contributor works with the specification file,
- spec-json-schemas where contributor pushes changes to JSON Schema of the spec,
- parser-js where contributor makes necessary changes in the JavaScript Parser.
The release coordinator should decide what the branch name needs to be, and contact the code owners for each repository to ask them to create the branches for them.
Once release branches are created, there are some initial changes that need to be made.
The process for doing this is the same for each of these:
- the release coordinator should create a fork of the relevant repository for these changes
- the commit message for the change should start with
chore: - the change should be contributed in a pull request targeting the release branch
- the release coordinator will need to ask the code owners for the relevant repository to approve and merge this pull request
Repository: spec
Examples are located in the examples/ folder in the spec repository. They should all be updated to the new version number.
An example of doing this is:
Repository: spec
References to the latest version in spec/asyncapi.md should be updated to the new version number.
Examples of doing this are:
Repository: spec-json-schemas
The new file should be created in the schemas/ folder in the spec-json-schemas repository.
It should be named with the version of the new release, and a link should be added to the index.js file (in the same repository).
An example of doing this is:
Repository: parser-js
The file to be updated is lib/asyncapiSchemaFormatParser.js in the parser-js repository.
The new version number should be added to the list in the getMimeTypes function.
An example of doing this is:
Once release branches have been updated, the default (e.g. "master") branches should be updated to identify the new release branch.
The process for doing this is the same for each of these:
- the release coordinator should create a fork of the relevant repository for these changes (this can be the same fork as used for updating the release branches)
- the commit message for the change should start with
chore: - the change should be contributed in a pull request targeting the default branch (normally
master) - the release coordinator will need to ask the code owners for the relevant repository to approve and merge this pull request
There are two repositories where package.json files need to be updated. In both repositories, the release branch name needs to be updated in the list of branches under .release.branches.
Examples of doing this are:
- this commit from the 2.3.0 release for parser-js
- this commit from the 2.3.0 release for spec-json-schemas
The release branch name needs to be updated in the .releaserc file in the spec repository needs.
An example of doing this is:
Each new release is announced by a blog post. You can see all of these at https://www.asyncapi.com/blog?tags=Release+Notes
The release coordinator should create an empty placeholder blog post that can be added to by different contributors to the release throughout the release process.
The steps to follow for this are:
- Create a fork of the website repository
- Create a new file at
pages/blog/release-notes-X.X.X.md(replacingX.X.Xwith the version number for the release) - Add a standard header at the top of the file (see the release notes for 2.2.0 and 2.3.0 for examples)
- Add a (webp format) cover image to the
public/img/posts/release-notes-X.X.X/folder, and update thecoverattribute of the blog post header with it. (Make sure to attribute the image correctly - unsplash.com is a good source of free images for this). See this commit from the 2.3.0 release for an example of doing this. - Add a (webp format) profile picture of the release coordinator to the
public/img/avatarsfolder, and update theauthorsattribute of the blog post header with it. See this commit from the 2.3.0 release for an example of doing this. - Open a draft pull request against the website repository. Make sure the option Allow edits and access to secrets by maintainers is selected to enable support from maintainers.
An example of doing this is:
Pull requests should be opened for all repositories covered by this process.
These should be full, not draft, pull requests to allow automated tests to run.
They should point from the release branches to the default/master branches for each repository.
Add a do-not-merge label to the pull request by making a comment in the PR saying /dnm.
Note: The automation bot will keep the release branch up to date with the latest commits from the master branch (so long as there are no conflicts).
The release coordinator should help to seek out possible updates that are good candidates for including in the release.
There are lots of ways to do this:
- ask for contributions in Slack
- ask for suggestions at a community meeting
- look for open accepted issues (see the contribution guide for a description of the requirements for a proposal to reach
Stage 3: Accepted) - look for open pull requests in the repositories covered by this process
For each feature that is being brought into the release, a pull request should be created from the feature branch (the branch with the accepted changes in) to the release branch.
Pull requests must be:
- labeled as an accepted proposal,
- created against the feature branch,
- created in all repositories specified in contribution guide
As features are identified for inclusion in the release, the draft announcement blog post should be updated with descriptions of them. The release cooordinator should coordinate with the feature contributors to write a description of each change. They should be able to provide input and also be allowed to work as co-authors for the release notes post.
Changes in the specification need to be well described. We need clear information on what has changed, why, and who contributed to the change. The purpose of the announcement blog post is to be a more user-friendly alternative to a regular changelog.
Every feature added to the release branch needs to be properly described in the release notes post.
In addition to the announcement blog post, the release coordinator should prepare release notes for each of the repositories covered by this process.
These should:
- be written in markdown
- include a list of features and significant changes
- include a link to github issues and pull requests where more background can be found
- include a link to a github diff with the previous release
Examples of doing this are:
As the release gets closer to being ready, it is helpful to remind the community that the release is on the way.
A good way to do this is to mention code owners and community tool maintainers in the release issue to draw their attention to it.
This can also be done:
- in the
#specificationchannel in Slack - at community meetings
Including a link to the release issue is a good way to let the community see the progress that has been made so far.
At least one code owner must approve the release pull requests in all related repositories.
Pre-release release candidates are generated automatically by the automation bot when:
- a pull request with a fix or feat prefix in the title is merged into the release branch
An example of a pull request created by the automation bot is: asyncapi/spec-json-schemas#151
Release candidates will include:
- upcoming release number
- date of the release
- release candidate number
An example release candidate is: https://github.com/asyncapi/spec/releases/tag/v2.3.0-2022-01-release.3
Important: When release candidates are created for the spec repository, the parser-js repository will need to be updated to use that release candidate.
Once everything is ready, it is time to merge the release branches using the draft pull requests prepared earlier.
Merging can only be done by code owners.
Pull requests must have a feat: prefix that, after the merge, produces a release candidate with a minor version update. The major version must have feat!: prefix.
Pull requests in the parser-js can be merged only if it uses the release candidate of @asyncapi/specs package produced after merge of a pull request in spec-json-schemas.
First, changes are merged into the spec repository, then spec-json-schemas and then in parser-js.
Release means merge of pull requests created from a release branch against the master branch. First, changes are merged into the spec repository, then spec-json-schemas and at the end in parser-js. Like in the case of the merge of release candidates, a pull request in parser-js can be merged only if it uses the final release of the new @asyncapi/specs package.
The release coordinator should ask the code owners for each repository to update the release in Github created by the automation bot, by adding the release notes they have prepared.
The release coordinator should notify maintainers of the following repositories that the first feature is merged and that release will be produced and therefore they need to start preparing for it:
The release coordinator should also make sure other maintainers from other projects under the AsyncAPI GitHub organization released their packages.
Every release of the release candidate is automatically published on the AsyncAPI Twitter account and in the releases-dedicated Slack channel.
If the release coordinator uses social networks like Twitter or LinkedIn, it is a great idea for them to promote the work that they've done to prepare the release by announcing it on their own private accounts. These can then be promoted and shared from the official AsyncAPI social accounts, to highlight and demonstrate the community-driven nature of releases.
Feel free to use other communication channels. Make sure that as many people as possible know about the change. Feel free to contact vendors upfront or other people that are interested in changes in the specification.
Every release identifies new issues and ways that the process can be improved.
Once the release is complete, the final step is to update this document with anything that could have gone better!