Release Process

This page aims to document our release process. It documents the release cycle, but also the steps required to make a release.

The release cycle

In general, each major Aegir release comprises a simultaneous release of all the modules that are part of the project. We generally go through several testing releases (alphas, betas & RCs) before doing the first stable release on a branch.

  • First an alpha is released to test new functionalities and to accomplish the goals decided in the project roadmap for that major version. Example: 3.0-alpha1
  • When we have covered most of the functionalities outlined in the roadmap, we push out beta releases until no more critical issues show up. This is generally considered a soft feature freeze. Example: 3.0-beta1
  • Then we go into full feature freeze and release a first release candidate (RC). Then a stable release branch is created, and the development branch is kept opened for development for the next stable release. This is generally considered a soft API freeze. Release candidates are made as long as critical bugs are found. Example: 3.0-rc1.
  • Once the development branch has no known critical bugs, the stable release is announced. From there on only critical fixes (security, critical performance and critical bugfixes) are committed to the stable branch, and stable releases are published (without alpha/beta/RC) directly on the stable branch. The stable branch is in full API freeze. New features are generally committed to the development branch.

Steps for a release

1. Make sure our tests are all green

Look into GitLab CI and Travis to see if all tasks have been performed without errors since the last commit. If there is an error, fix it before the release.

2. Verify drupal-org.make

In the hostmaster project we maintain our own drupal-org.make file. Verify that drupal-org.make specifies up-to-date versions. Check that e.g. the ctools version specified is not out-dated.

3. Generating the release notes

We build complete release notes for every release. Those are made up of a summary of the release, an outline of key changes, of known issues, install and upgrade instructions and a full list of bugfixes and new features. As we have a number of Drupal.org projects to cover we try to centralize our release notes combined with the documentation. Published under the release notes section. On Drupal.org we add a link to the version specific release notes page on the release nodes for all projects we cover.

Using Git Release Notes for Drush with a patch for the MD format:

drush rn --baseurl=https://www.drupal.org/ --md 7.x-3.10 HEAD

or after placing the tags

drush rn --baseurl=https://www.drupal.org/ --md 7.x-3.3 7.x-3.4

This is done for the different git projects. Changing the first line to add 'to $name'. The developers then proceed to format/edit the list of fixes as well as list other significant information/changes for this release. These notes end up becoming the Release Notes for the release.

Add it as a new file under release-notes, and add a line to mkdocs.yml. And update the default page for the 'Release notes' section.

4. Running the release.sh script

Each time we make a new release, we run a script called release.sh in provision. This script does all the 'hard' work in that it doesn't forget all the very many places to edit version numbers etc of relevant documentation and other scripts. This includes updating upgrade.sh.txt.

It's best to start in a freshy cloned provision checkout. An extra remote to GitLab is needed for the packaging by GitLab CI.

git clone $USER@git.drupal.org:project/provision.git
cd provision/
git remote add gitlab git@gitlab.com:aegir/provision.git

Paraphrasing from the script itself:

$ ./scripts/release.sh 3.11

Aegir release script
====================

This script should only be used by the core dev team when doing an
official release. If you are not one of those people, you probably
shouldn't be running this.

This script is going to modify the configs and documentation to
release 7.x-3.7.

The following operations will be done:
 0. prompt you for a debian/changelog entry
 1. change the makefile to download tarball
 2. change the upgrade.sh.txt version
 3. change the .gitlab-ci.yml to build a release package
 4. display the resulting diff
 5. commit those changes to git
 6. lay down the tag
 7. revert the commit
 8. clone fresh copies of hosting/hostmaster and eldir to lay down the tag
 9. (optionally) push those changes
10. clone fresh copies of golden contrib to lay down the tag
11. (optionally) push those changes

The operation can be aborted before step 8. Don't forget that as
long as changes are not pushed upstream, this can all be reverted (see
git-reset(1) and git-revert(1) ).

Notice how we just provide the Aegir release number (3.11) to the release script, not the Drupal branch (7.x), which is hardcoded in the script to remove potential confusion.

5. Watch the GitLab CI pipelines

After the release script pushes it's commits you should give Ci some time to finish it's tests.

If any of these builds fail, one option might be to delete the remote tags (using git push origin :7.x-3.4, for example), fix the bugs and start again.

6 Fix Debian packages only

When there's a bug in the Debian packaging itself we can do a minor package version update. USE WITH CARE!

cd <provision source>
<Fix>
git commit
git tag <previous version tag>.1
git push --tags

TODO: We also need to update version numbers(as release.sh does), but only for provision. Maybe the script can get an option for it...

We don't create a release node on Drupal.org, just let CI create the packages and publish them as described below.

7. Creating release nodes on Drupal.org

Once the tags are pushed and release notes published, we create a release node with an excerpt of (and a link to) the release notes so that tarballs are created and issue queue versions updated.

Using the template (change the version number):

See the full release notes at: http://docs.aegirproject.org/en/3.x/release-notes/3.11/

This needs to be done in the provision and hosting and (maybe) eldir projects on Drupal.org. And for Golden Contrib... hosting_civicrm, hosting_git, hosting_remote_import, hosting_site_backup_manager and hosting_tasks_extra

WAIT.... And only after those are fully build... in the hostmaster project.

Note: this could be automated with the right stuff on Drupal.org.

9. Manually test install and upgrade

Just to make sure you should test the install and upgrade to the new version in a local VM. (Vagrant is very useful for this, the provision repo has a Vagrantfile.)

At this point, the unstable repo actually contains the future stable version (ie 3.1 instead of 3.1-dev-abc). Check that it actually does: http://debian.aegirproject.org/dists/unstable/main/binary-amd64/Packages

To test the install, just use the instructions on http://www.aegirproject.org but replace "stable" by "unstable". You should end up with a stable version.

To test the upgrade, create a new VM, install normally with the aegirproject.org instructions, then replace stable by unstable in your lists (/etc/apt/), and run

apt-get update && apt-get upgrade

It should ask you if you want to upgrade Aegir3. Say yes and make sure there are no errors.

If you don't encounter errors in this procedure, you're good to go.

9.1. Quick test VMs

You can use an Vagrant base box to quickly test the install and upgrade. Some such boxes are optimized for Aegir development and testing:

(Please add your base box to the list if relevant.)

10. Publish the Debian packages

Finally, when the Debian packages are tested you will need to pull them into the stable release channel:

We pull to stable (since the betas), manually:

ssh aegir0.aegir.coop
sudo su - reprepro
reprepro@aegir0:~$
reprepro copy stable unstable aegir3
reprepro copy stable unstable aegir3-hostmaster
reprepro copy stable unstable aegir3-provision
reprepro copy stable unstable aegir3-cluster-slave

If CI has managed to build .debs and upload them before you've have a chance to pull them into testing/stable, you can manually remove them like so:

reprepro remove unstable aegir3-cluster-slave aegir3 aegir3-provision aegir3-hostmaster

You can then re-upload the new .debs you've generated, using the '-f' (force) flag:

dput -f aegir build-area/aegir2-provision_2.3_amd64.changes

After doing that, you can re-run the 'copy' commands to publish the .debs to the appropriate releases.

11. Publish the release notes widely

Once all this is done and the tarballs are generated, the release notes are published in:

  • The docs
  • Update the upgrade.sh link in the documentation
  • The topic of the IRC channel
  • The aegir-announce mailing list (announce@lists.aegirproject.org.)
  • Twitter as @aegirproject
    • Template: "#aegir 3.? released! Release notes at http://docs.aegirproject.org/en/3.x/release-notes/3.?/"

Optionally, blog posts on koumbit.org, mig5.net, and elsewhere may go into further detail about significant changes, screencasts etc.

12. Debian repository management

The repo is managed with a tool called reprepro. Gitlab scp's to /var/www/repos/incoming/ on the repo server. It has an ssh key stored as a secret variable.

On the repo server incrond waits for new files and trigers reprepro using /usr/local/bin/reprepro_process_incomming for them.

Setting up a GitLab runner

GitLab.com offers the use of shares runner VM's to use for CI, but we need a priviledged one to be able to use sudo and sub docker images. As a project we maintain one runner for this, but if you want to experiment more in a feature branch of your own fork then it's best to setup a new one. Documentation on setting one up is on https://docs.gitlab.com/runner/install/ The minimal memory requirement seems to be 1GB. And up to 10GB of disk storage.

Example config file /etc/gitlab-runner/config.toml:

concurrent = 1
check_interval = 0

[[runners]]
  name = "Initfour-helmo-ci_runner01"
  url = "https://gitlab.com/ci"
  token = "9N-J78KC9b7oQE8LMP_d"
  executor = "docker"
  [runners.docker]
    tls_verify = false
    image = "ruby:2.1"
    privileged = true
    disable_cache = false
    volumes = ["/var/run/docker.sock:/var/run/docker.sock", "/cache"]
  [runners.cache]

The modified lines are 'privileged' and 'volumes'

Troubleshooting CI tests

When you have your own GitLab CI runner you can inspect the container while it's running the tests. Adding an extra sleep 20m as a last task in the .gitlab-ci.yml script section would keep it alive a bit longer. The default timeout is 3600 seconds.

Use docker ps to get the container ID, and docker exec -ti [containerID] bash to get a shell inside the container.