We have three channels to which we can release: production
, beta
, and test
.
-
production
is the channel from which the general public downloads and receives updates. It should be stable and polished. -
beta
is released more often thanproduction
. We want to ensuredevelopment
is always in a state where it can be released to users, so it should be used as the source forbeta
releases as an opportunity for additional QA before releasing toproduction
. -
test
is unlike the other two. It does not receive updates. Each test release is locked in time. It's used entirely for providing test releases.
From a clean working directory, set the GITHUB_ACCESS_TOKEN
environment variable to a valid Personal Access Token
To check that this environment variable is set in your shell:
Bash (macOS, Linux or Git Bash)
$ echo $GITHUB_ACCESS_TOKEN
Command Prompt
$ echo %GITHUB_ACCESS_TOKEN%
PowerShell
$ echo $env:GITHUB_ACCESS_TOKEN
If you are creating a new Personal Access Token on GitHub:
- make the token memorable - use a description like
Desktop Draft Release and Changelog Generator
- the
read:org
scope is the only required scope for drafting releases
To set this access token as an environment in your shell:
Bash (macOS, Linux or Git Bash)
$ export GITHUB_ACCESS_TOKEN={your token here}
Command Prompt
$ set GITHUB_ACCESS_TOKEN={your token here}
PowerShell
$ $env:GITHUB_ACCESS_TOKEN="{your token here}"
Create a new branch to represent the work that will be released to users:
- for
beta
releases, branch fromdevelopment
to ensure the latest changes are published - for
production
releases, branch from the latest beta tag- to find this tag:
git tag | grep 'beta' | sort -r | head -n 1
- to find this tag:
When naming the branch, ensure you use the releases/[version]
pattern to ensure all CI platforms are aware of the branch and will build any PRs that target the branch.
Run the script below (which relies on the your personal acccess token being set), which will determine the next version from what was previously published, based on the desired channel.
For production
and beta
releases, run:
$ yarn draft-release (production|beta)
If you are creating a new beta release, the yarn draft-release beta
command will help you find the new release entries for the changelog.
If you are create a new production
release, you should just combine and sort the previous beta
changelog entries.
(For test
releases, follow the directions in the steps below to update app/package.json
's version
to a higher version and add a changelog entry. The script does not support test yet.)
The script will output a draft changelog, which covers everything that's been merged, and probably needs some love. The output will then explain the next steps:
Here's what you should do next:
1. Update the app/package.json 'version' to '1.0.14-beta2' (make sure this aligns with semver format of 'major.minor.patch')
2. Concatenate this to the beginning of the releases element in the changelog.json as a starting point:
{
"1.0.14-beta2": [
"[???] Add RubyMine support for macOS - #3883. Thanks @gssbzn!",
"[???] Allow window to accept single click on focus - #3843",
"[???] Drop unnecessary comments before issue template - #3906",
"[???] First-class changelog script for generating release notes - #3888",
"[???] Fix expanded avatar stack overflow - #3884",
"[???] Switch to a saner default gravatar size - #3911",
"[Fixed] Add a repository settings store - #934",
"[Fixed] Ensure renames are detected when viewing commit diffs - #3673",
"[Fixed] Line endings are hard, lets go shopping - #3514",
]
}
3. Revise the release notes according to https://github.com/desktop/desktop/blob/development/docs/process/writing-release-notes.md
4. Commit the changes (on development or as new branch) and push them to GitHub
5. Read this to perform the release: https://github.com/desktop/desktop/blob/development/docs/process/releasing-updates.md
See our release notes writing guide for more info on how we write and review our release notes.
Note: You should ensure the version
in app/package.json
is set to the new version and follows the semver format of major.minor.patch
.
Examples:
- for prod,
1.1.0
->1.1.1
or1.1.13
->1.2.0
- for beta,
1.1.0-beta1
->1.1.0-beta2
or1.1.13-beta3
->1.2.0-beta1
- for test,
1.0.14-test2
->1.0.14-test3
or1.1.14-test3
->1.2.0-test1
Here's an example of the previous changelog draft after it has been edited:
{
"1.0.14-beta2": [
"[Added] Add RubyMine support for macOS - #3883. Thanks @gssbzn!",
"[Fixed] Allow window to accept single click on focus - #3843",
"[Fixed] Expanded avatar list hidden behind commit details - #3884",
"[Fixed] Renames not detected when viewing commit diffs - #3673",
"[Fixed] Ignore action assumes CRLF when core.autocrlf is unset - #3514",
"[Improved] Use smaller default size when rendering Gravatar avatars - #3911",
]
}
Add your new changelog entries to changelog.json
, update the version in app/package.json
, commit the changes, and push this branch to GitHub. This becomes the release branch, and lets other maintainers continue to merge into development
without affecting your release.
If a maintainer would like to backport a pull request to the next release, it is their responsibilty to co-ordinate with the release owner and ensure they are fine with accepting this work.
Once your release branch is ready to review and ship, ask the other maintainers to review and approve the changes!
IMPORTANT NOTE: Do NOT "Update branch" and merge development into the release branch. This might be tempting if the "branch is out-of-date with the base branch" dotcom feature is enabled. However, doing so would inadvertently release everything on development to production or beta 🙀
When you are ready to start the deployment, run this command in chat (where X.Y.Z-release
is the name of your release branch):
.release! desktop/X.Y.Z-release to {production|beta|test}
We're using .release
with a bang so that we don't have to wait for any current CI on the branch to finish. This might feel a little wrong, but it's OK since making the release itself will also run CI.
If you're releasing a production
update, release a beta
update for the next version too, so that beta users are on the latest release. For example, if the version just released to production is 1.2.0
then the beta release should be 1.2.1-beta0
to indicate there are no new changes on top of what's currently on production
.
IMPORTANT NOTE: Ensure that you indicate which channel to release to. If not, chatops will default to releasing to production 🙀
Go to Central's Deployments to find your release; you'll see something at the top of the page like:
desktop/desktop deployed from {YOUR_BRANCH}@{HASH_ABBREVIATION_FOR_COMMIT} to {production|beta|test}
it will initially specify its state as State: pending
and will be completed when it says State: released
You will also see this in Chat:
desktopbot tagged desktop/release-{YOUR_VERSION}
When the release in Central is in State: released
for beta
or production
, switch to your installed Desktop instance and make sure that the corresponding (prod|beta) app auto-updates.
Testing that an update is detected, downloaded, and applied correctly is very important - if this is somehow broken during development then our users will not likely stay up to date!
If you don't have the app for beta
, for example, you can always download the previous version on Central to see it update
Make sure you move your application out of the Downloads folder and into the Applications folder for macOS or it won't auto-update.
So that we keep the changelog.json
up to date. Beta entries will be used for the upcoming production release.
If an error occurs during the release process, a needle will be reported to Central's Haystack.
After the release is deployed, you should monitor Desktop's Haystack closely for 15 minutes to ensure no unexpected needles appear.
If the active beta is the last beta prior to a production release, extra care should be taken when looking at Desktop's Haystack roll-ups. The lead engineer responsible for deployment should produce a Haystack report the day before and after the release. The report should contain a list of any new or unexpected errors from the past beta releases in the milestone and be published to the team's Slack channel.
Once your app updates and you see the visible changes in your app and there are no spikes in errors, celebrate 🎉!!! You did it!
Also it might make sense to continue to monitor Haystack in the background for the next 24 hours.
So let's say you kicked off a release with chatops on accident. Here's how you fix that.
When you kicked off the release, a branch with the prefix __release-${channel}-
was created in the GitHub repo. Use that branch name to find the proper CI jobs below.
- Delete the pending release from Central
- Cancel the Appveyor release job
- Cancel the CircleCI release job
- Delete the CI release job branch from GitHub
- Breathe a sigh of relief
You don't need to do anything with your manually created release branch, that you referred to in the chatops command. Feel free to re-use it.