docker/project/RELEASE-CHECKLIST.md

17 KiB

Release Checklist

A maintainer's guide to releasing Docker

So you're in charge of a Docker release? Cool. Here's what to do.

If your experience deviates from this document, please document the changes to keep it up-to-date.

It is important to note that this document assumes that the git remote in your repository that corresponds to "https://github.com/docker/docker" is named "origin". If yours is not (for example, if you've chosen to name it "upstream" or something similar instead), be sure to adjust the listed snippets for your local environment accordingly. If you are not sure what your upstream remote is named, use a command like git remote -v to find out.

If you don't have an upstream remote, you can add one easily using something like:

export GITHUBUSER="YOUR_GITHUB_USER"
git remote add origin https://github.com/docker/docker.git
git remote add $GITHUBUSER git@github.com:$GITHUBUSER/docker.git

1. Pull from master and create a release branch

All releases version numbers will be of the form: vX.Y.Z where X is the major version number, Y is the minor version number and Z is the patch release version number.

Major releases

The release branch name is just vX.Y because it's going to be the basis for all .Z releases.

export BASE=vX.Y
export VERSION=vX.Y.Z
git fetch origin
git checkout --track origin/master
git checkout -b release/$BASE

This new branch is going to be the base for the release. We need to push it to origin so we can track the cherry-picked changes and the version bump:

git push origin release/$BASE

When you have the major release branch in origin, we need to create the bump fork branch that we'll push to our fork:

git checkout -b bump_$VERSION

Patch releases

If we have the release branch in origin, we can create the forked bump branch from it directly:

export VERSION=vX.Y.Z
export PATCH=vX.Y.Z+1
git fetch origin
git checkout --track origin/release/$BASE
git checkout -b bump_$PATCH

We cherry-pick only the commits we want into the bump branch:

# get the commits ids we want to cherry-pick
git log
# cherry-pick the commits starting from the oldest one, without including merge commits
git cherry-pick -s -x <commit-id>
git cherry-pick -s -x <commit-id>
...

2. Update the VERSION files and API version on master

We don't want to stop contributions to master just because we are releasing. So, after the release branch is up, we bump the VERSION and API version to mark the start of the "next" release.

2.1 Update the VERSION files

Update the content of the VERSION file to be the next minor (incrementing Y) and add the -dev suffix. For example, after the release branch for 1.5.0 is created, the VERSION file gets updated to 1.6.0-dev (as in "1.6.0 in the making").

2.2 Update API version on master

We don't want API changes to go to the now frozen API version. Create a new entry in docs/reference/api/ by copying the latest and bumping the version number (in both the file's name and content), and submit this in a PR against master.

3. Update CHANGELOG.md

You can run this command for reference with git 2.0:

git fetch --tags
LAST_VERSION=$(git tag -l --sort=-version:refname "v*" | grep -E 'v[0-9\.]+$' | head -1)
git log --stat $LAST_VERSION..bump_$VERSION

If you don't have git 2.0 but have a sort command that supports -V:

git fetch --tags
LAST_VERSION=$(git tag -l | grep -E 'v[0-9\.]+$' | sort -rV | head -1)
git log --stat $LAST_VERSION..bump_$VERSION

If releasing a major version (X or Y increased in vX.Y.Z), simply listing notable user-facing features is sufficient.

#### Notable features since <last major version>
* New docker command to do something useful
* Engine API change (deprecating old version)
* Performance improvements in some usecases
* ...

For minor releases (only Z increases in vX.Y.Z), provide a list of user-facing changes. Each change should be listed under a category heading formatted as #### CATEGORY.

CATEGORY should describe which part of the project is affected. Valid categories are:

  • Builder
  • Documentation
  • Hack
  • Packaging
  • Engine API
  • Runtime
  • Other (please use this category sparingly)

Each change should be formatted as BULLET DESCRIPTION, given:

  • BULLET: either -, + or *, to indicate a bugfix, new feature or upgrade, respectively.

  • DESCRIPTION: a concise description of the change that is relevant to the end-user, using the present tense. Changes should be described in terms of how they affect the user, for example "Add new feature X which allows Y", "Fix bug which caused X", "Increase performance of Y".

EXAMPLES:

## 0.3.6 (1995-12-25)

#### Builder

+ 'docker build -t FOO .' applies the tag FOO to the newly built image

#### Engine API

- Fix a bug in the optional unix socket transport

#### Runtime

* Improve detection of kernel version

If you need a list of contributors between the last major release and the current bump branch, use something like:

git log --format='%aN <%aE>' v0.7.0...bump_v0.8.0 | sort -uf

Obviously, you'll need to adjust version numbers as necessary. If you just need a count, add a simple | wc -l.

4. Change the contents of the VERSION file

Before the big thing, you'll want to make successive release candidates and get people to test. The release candidate number N should be part of the version:

export RC_VERSION=${VERSION}-rcN
echo ${RC_VERSION#v} > VERSION

5. Test the docs

Make sure that your tree includes documentation for any modified or new features, syntax or semantic changes.

To test locally:

make docs

To make a shared test at https://beta-docs.docker.io:

(You will need the awsconfig file added to the docs/ dir)

make AWS_S3_BUCKET=beta-docs.docker.io BUILD_ROOT=yes docs-release

6. Commit and create a pull request to the "release" branch

git add VERSION CHANGELOG.md
git commit -m "Bump version to $VERSION"
git push $GITHUBUSER bump_$VERSION
echo "https://github.com/$GITHUBUSER/docker/compare/docker:release/$BASE...$GITHUBUSER:bump_$VERSION?expand=1"

That last command will give you the proper link to visit to ensure that you open the PR against the "release" branch instead of accidentally against "master" (like so many brave souls before you already have).

7. Create a PR to update the AUTHORS file for the release

Update the AUTHORS file, by running the hack/generate-authors.sh on the release branch. To prevent duplicate entries, you may need to update the .mailmap file accordingly.

8. Build release candidate rpms and debs

NOTE: It will be a lot faster if you pass a different graphdriver with DOCKER_GRAPHDRIVER than vfs.

docker build -t docker .
docker run \
    --rm -t --privileged \
    -e DOCKER_GRAPHDRIVER=aufs \
    -v $(pwd)/bundles:/go/src/github.com/docker/docker/bundles \
    docker \
    hack/make.sh binary build-deb build-rpm

9. Publish release candidate rpms and debs

With the rpms and debs you built from the last step you can release them on the same server, or ideally, move them to a dedicated release box via scp into another docker/docker directory in bundles. This next step assumes you have a checkout of the docker source code at the same commit you used to build, with the artifacts from the last step in bundles.

NOTE: If you put a space before the command your .bash_history will not save it. (for the GPG_PASSPHRASE).

docker build -t docker .
docker run --rm -it --privileged \
    -v /volumes/repos:/volumes/repos \
    -v $(pwd)/bundles:/go/src/github.com/docker/docker/bundles \
    -v $HOME/.gnupg:/root/.gnupg \
    -e DOCKER_RELEASE_DIR=/volumes/repos \
    -e GPG_PASSPHRASE \
    -e KEEPBUNDLE=1 \
    docker \
    hack/make.sh release-deb release-rpm sign-repos generate-index-listing

10. Upload the changed repos to wherever you host

For example, above we bind mounted /volumes/repos as the storage for DOCKER_RELEASE_DIR. In this case /volumes/repos/apt can be synced with a specific s3 bucket for the apt repo and /volumes/repos/yum can be synced with a s3 bucket for the yum repo.

11. Publish release candidate binaries

To run this you will need access to the release credentials. Get them from the Core maintainers.

docker build -t docker .

# static binaries are still pushed to s3
docker run \
    -e AWS_S3_BUCKET=test.docker.com \
    -e AWS_ACCESS_KEY_ID \
    -e AWS_SECRET_ACCESS_KEY \
    -e AWS_DEFAULT_REGION \
    -i -t --privileged \
    docker \
    hack/release.sh

It will run the test suite, build the binaries and upload to the specified bucket, so this is a good time to verify that you're running against test.docker.com.

12. Purge the cache!

After the binaries are uploaded to test.docker.com and the packages are on apt.dockerproject.org and yum.dockerproject.org, make sure they get tested in both Ubuntu and Debian for any obvious installation issues or runtime issues.

If everything looks good, it's time to create a git tag for this candidate:

git tag -a $RC_VERSION -m $RC_VERSION bump_$VERSION
git push origin $RC_VERSION

Announcing on multiple medias is the best way to get some help testing! An easy way to get some useful links for sharing:

echo "Ubuntu/Debian: curl -sSL https://test.docker.com/ | sh"
echo "Linux 64bit binary: https://test.docker.com/builds/Linux/x86_64/docker-${VERSION#v}"
echo "Darwin/OSX 64bit client binary: https://test.docker.com/builds/Darwin/x86_64/docker-${VERSION#v}"
echo "Linux 64bit tgz: https://test.docker.com/builds/Linux/x86_64/docker-${VERSION#v}.tgz"
echo "Windows 64bit client binary: https://test.docker.com/builds/Windows/x86_64/docker-${VERSION#v}.exe"
echo "Windows 32bit client binary: https://test.docker.com/builds/Windows/i386/docker-${VERSION#v}.exe"

We recommend announcing the release candidate on:

  • IRC on #docker, #docker-dev, #docker-maintainers
  • In a comment on the pull request to notify subscribed people on GitHub
  • The docker-dev group
  • The docker-maintainers group
  • Any social media that can bring some attention to the release candidate

13. Iterate on successive release candidates

Spend several days along with the community explicitly investing time and resources to try and break Docker in every possible way, documenting any findings pertinent to the release. This time should be spent testing and finding ways in which the release might have caused various features or upgrade environments to have issues, not coding. During this time, the release is in code freeze, and any additional code changes will be pushed out to the next release.

It should include various levels of breaking Docker, beyond just using Docker by the book.

Any issues found may still remain issues for this release, but they should be documented and give appropriate warnings.

During this phase, the bump_$VERSION branch will keep evolving as you will produce new release candidates. The frequency of new candidates is up to the release manager: use your best judgement taking into account the severity of reported issues, testers availability, and time to scheduled release date.

Each time you'll want to produce a new release candidate, you will start by adding commits to the branch, usually by cherry-picking from master:

git cherry-pick -s -x -m0 <commit_id>

You want your "bump commit" (the one that updates the CHANGELOG and VERSION files) to remain on top, so you'll have to git rebase -i to bring it back up.

Now that your bump commit is back on top, you will need to update the CHANGELOG file (if appropriate for this particular release candidate), and update the VERSION file to increment the RC number:

export RC_VERSION=$VERSION-rcN
echo $RC_VERSION > VERSION

You can now amend your last commit and update the bump branch:

git commit --amend
git push -f $GITHUBUSER bump_$VERSION

Repeat step 6 to tag the code, publish new binaries, announce availability, and get help testing.

14. Finalize the bump branch

When you're happy with the quality of a release candidate, you can move on and create the real thing.

You will first have to amend the "bump commit" to drop the release candidate suffix in the VERSION file:

echo $VERSION > VERSION
git add VERSION
git commit --amend

You will then repeat step 6 to publish the binaries to test

15. Get 2 other maintainers to validate the pull request

16. Build final rpms and debs

docker build -t docker .
docker run \
    --rm -t --privileged \
    -v $(pwd)/bundles:/go/src/github.com/docker/docker/bundles \
    docker \
    hack/make.sh binary build-deb build-rpm

17. Publish final rpms and debs

With the rpms and debs you built from the last step you can release them on the same server, or ideally, move them to a dedicated release box via scp into another docker/docker directory in bundles. This next step assumes you have a checkout of the docker source code at the same commit you used to build, with the artifacts from the last step in bundles.

NOTE: If you put a space before the command your .bash_history will not save it. (for the GPG_PASSPHRASE).

docker build -t docker .
docker run --rm -it --privileged \
    -v /volumes/repos:/volumes/repos \
    -v $(pwd)/bundles:/go/src/github.com/docker/docker/bundles \
    -v $HOME/.gnupg:/root/.gnupg \
    -e DOCKER_RELEASE_DIR=/volumes/repos \
    -e GPG_PASSPHRASE \
    -e KEEPBUNDLE=1 \
    docker \
    hack/make.sh release-deb release-rpm sign-repos generate-index-listing

18. Upload the changed repos to wherever you host

For example, above we bind mounted /volumes/repos as the storage for DOCKER_RELEASE_DIR. In this case /volumes/repos/apt can be synced with a specific s3 bucket for the apt repo and /volumes/repos/yum can be synced with a s3 bucket for the yum repo.

19. Publish final binaries

Once they're tested and reasonably believed to be working, run against get.docker.com:

docker build -t docker .
# static binaries are still pushed to s3
docker run \
    -e AWS_S3_BUCKET=get.docker.com \
    -e AWS_ACCESS_KEY_ID \
    -e AWS_SECRET_ACCESS_KEY \
    -e AWS_DEFAULT_REGION \
    -i -t --privileged \
    docker \
    hack/release.sh

20. Purge the cache!

21. Apply tag and create release

It's very important that we don't make the tag until after the official release is uploaded to get.docker.com!

git tag -a $VERSION -m $VERSION bump_$VERSION
git push origin $VERSION

Once the tag is pushed, go to GitHub and create a new release. If the tag is for an RC make sure you check This is a pre-release at the bottom of the form.

Select the tag that you just pushed as the version and paste the changelog in the description of the release. You can see examples in this two links:

https://github.com/docker/docker/releases/tag/v1.8.0 https://github.com/docker/docker/releases/tag/v1.8.0-rc3

22. Go to github to merge the bump_$VERSION branch into release

Don't forget to push that pretty blue button to delete the leftover branch afterwards!

23. Update the docs branch

You will need to point the docs branch to the newly created release tag:

git checkout origin/docs
git reset --hard origin/$VERSION
git push -f origin docs

The docs will appear on https://docs.docker.com/ (though there may be cached versions, so its worth checking http://docs.docker.com.s3-website-us-east-1.amazonaws.com/). For more information about documentation releases, see docs/README.md.

Note that the new docs will not appear live on the site until the cache (a complex, distributed CDN system) is flushed. The make docs-release command will do this if the DISTRIBUTION_ID is set correctly - this will take at least 15 minutes to run and you can check its progress with the CDN Cloudfront Chrome addon.

24. Create a new pull request to merge your bump commit back into master

git checkout master
git fetch
git reset --hard origin/master
git cherry-pick -s -x $VERSION
git push $GITHUBUSER merge_release_$VERSION
echo "https://github.com/$GITHUBUSER/docker/compare/docker:master...$GITHUBUSER:merge_release_$VERSION?expand=1"

Again, get two maintainers to validate, then merge, then push that pretty blue button to delete your branch.

25. Rejoice and Evangelize!

Congratulations! You're done.

Go forth and announce the glad tidings of the new release in #docker, #docker-dev, on the dev mailing list, the announce mailing list, and on Twitter!