In this document, we’ll cover the steps necessary to release Rails. Each section contains steps to take during that time before the release. The times suggested in each header are just that: suggestions. However, they should really be considered as minimums.
Today is mostly coordination tasks. Here are the things you must do today:
Do not release with a Red CI. You can find the CI status here:
http://travis-ci.org/rails/rails
Sam Ruby keeps a test suite that makes sure the code samples in his book (Agile Web Development with Rails) all work. These are valuable integration tests for Rails. You can check the status of his tests here:
http://intertwingly.net/projects/dashboard.html
Do not release with Red AWDwR tests.
Having Git dependencies indicates that we depend on unreleased code. Obviously Rails cannot be released when it depends on unreleased code. Contact the authors of those particular gems and work out a release date that suits them.
Let them know of your plans to release. There may be security issues to be addressed, and that can impact your release date.
Ruby implementors have high stakes in making sure Rails works. Be kind and give them a heads up that Rails will be released soonish.
This only needs done for major and minor releases, bugfix releases aren’t a big enough deal, and are supposed to be backwards compatible.
Send an email just giving a heads up about the upcoming release to these lists:
Implementors will love you and help you.
This is when you should release the release candidate. Here are your tasks for today:
From the stable branch, create a release branch. For example, if you’re releasing Rails 3.0.10, do this:
[aaron@higgins rails (3-0-stable)]$ git checkout -b 3-0-10 Switched to a new branch '3-0-10' [aaron@higgins rails (3-0-10)]$
Many times commits are made without the CHANGELOG being updated. You should review the commits since the last release, and fill in any missing information for each CHANGELOG.
You can review the commits for the 3.0.10 release like this:
[aaron@higgins rails (3-0-10)]$ git log v3.0.9..
If you’re doing a stable branch release, you should also ensure that all of the CHANGELOG entries in the stable branch are also synced to the master branch.
Run ‘rake install` to generate the gems and install them locally. Then try generating a new app and ensure that nothing explodes.
This will stop you from looking silly when you push an RC to rubygems.org and then realise it is broken.
IMPORTANT: Due to YAML parse problems on the rubygems.org server, it is safest to use Ruby 1.8 when releasing.
Run ‘rake release`. This will populate the gemspecs with data from RAILS_VERSION, commit the changes, tag it, and push the gems to rubygems.org. Here are the commands that `rake release` should use, so you can understand what to do in case anything goes wrong:
$ rake all:build $ git commit -am'updating RAILS_VERSION' $ git tag -m'tagging rc release' v3.0.10.rc1 $ git push $ git push --tags $ for i in $(ls pkg); do gem push $i; done
Write a release announcement that includes the version, changes, and links to GitHub where people can find the specific commit list. Here are the mailing lists where you should announce:
Use Markdown format for your announcement. Remember to ask people to report issues with the release candidate to the rails-core mailing list.
IMPORTANT: If any users experience regressions when using the release candidate, you must postpone the release. Bugfix releases *should not* break existing applications.
If you used Markdown format for your email, you can just paste it in to the blog.
Check the rails-core mailing list and the GitHub issue list for regressions in the RC.
If any regressions are found, fix the regressions and repeat the release candidate process. We will not release the final until 72 hours after the last release candidate has been pushed. This means that if users find regressions, the scheduled release date must be postponed.
When you fix the regressions, do not create a new branch. Fix them on the stable branch, then cherry pick the commit to your release branch. No other commits should be added to the release branch besides regression fixing commits.
Many of these steps are the same as for the release candidate, so if you need more explanation on a particular step, so the RC steps.
Today, do this stuff in this order:
-
Apply security patches to the release branch
-
Update CHANGELOG with security fixes.
-
Update RAILS_VERSION to remove the rc
-
Build and test the gem
-
Release the gems
-
If releasing a new stable version:
-
Trigger stable docs generation (see below)
-
Update the version in the home page
-
-
Email security lists
-
Email general announcement lists
Email the security announce list once for each vulnerability fixed.
You can do this, or ask the security team to do it.
Email the security reports to:
Be sure to note the security fixes in your announcement along with CVE numbers and links to each patch. Some people may not be able to upgrade right away, so we need to give them the security fixes in patch form.
-
Blog announcements
-
Twitter announcements
-
Merge the release branch to the stable branch.
-
Drink beer (or other cocktail)
There are two simple steps for fixing the CI:
-
Identify the problem
-
Fix it
Repeat these steps until the CI is green.
We have a post-receive hook in GitHub that calls the docs server on pushes. It triggers generation and publication of edge docs, updates the contrib app, and generates and publishes stable docs if a new stable tag is detected.
The hook unfortunately is not invoked by tag pushing, so once the new stable tag has been pushed to origin, please run
rake publish_docs
You should see something like this:
Rails master hook tasks scheduled: * updates the local checkout * updates Rails Contributors * generates and publishes edge docs If a new stable tag is detected it also * generates and publishes stable docs This needs typically a few minutes.
Note you do not need to specify the tag, the docs server figures it out.
Also, don’t worry if you call that multiple times or the hook is triggered again by some immediate regular push, if the scripts are running new calls are just queued (in a queue of size 1).