Firefox Release Process¶
Requirements and permissions to conduct a release¶
In order to initiate and ship a Firefox release, follow these steps to be granted the various permissions required.
Specific steps required to initiate and release Firefox¶
Once you have the permissions to conduct a release, the following is a walkthrough of how to do it.
Connect to Shipit.
Initiate a new release
Trigger the various phases of the release
Sign off in Balrog
Connect to the VPN¶
Connect to the VPN. Visit https://shipit.mozilla-releng.net/. Sign in using your LDAP credentials.
Initiate a new release¶
From Shipit, click on “New Release”. Choose the target product, e.g. Firefox Desktop. Choose the target channel, e.g. Beta, and select the desired revision you would like to ship from.
Revision - you must use a valid revision from the chosen channel. e.g. if you selected Beta, you must choose a revision from mozilla-beta.
Release ETA - If you are asked to give a “Release ETA”, you do not need to fill this out.
Partials - partials should be auto populated. The list of partials are previous released versions that users are using currently that we want to provide a special update to that is small in download size and is a diff of the user’s current version and the new version of Firefox you are about to create and ship. The auto-populated versions are usually based on which versions are currently the most used by users.
A note on release promotion: Releases do not create new builds of Firefox. Instead, the automation will take existing builds from the revision that was created and built when that revision was pushed (checked in). We refer to this as “Release Promotion”. For that reason, the revision must have builds started or complete and associated with the target revision. To see the builds of a given pushed revision, use Treeherder.
Version and build number - both of these are auto populated and can not be modified. The build number refers to how many times we have had to create and recreate a release prior to shipping it to users. This often happens when the release automation fails or or Firefox fails to pass QA.
Once the form is filled out, click Start tracking it. Note this does not actually kick off any release automation. It merely primes the release in Shipit so that you can start triggering the various phases.
For more on how to create a release in Shipit, including UI screenshots, see Relman’s documentation
Trigger the phases of the release¶
After creating a release in Shipit, you can see it via the “Releases” tab. From that tab, you can initiate a phase. For most products, there are three phases: “promote”, “push”, and “ship”.
Promote - as mentioned above, Shipit and release automation do not actually create any new Firefox builds. It instead takes existing per-checkin builds from CI and signs, repacks, and publishes those to users. The promote phase consists of release automation that does just that. Common tasks in this phase are signing, creating localization specific builds, uploading build artifacts to a staging location on http://archive.mozilla.org/ (S3), staging the release on our update server, Balrog, and running install and update testing.
Push - This phase takes all the build artifacts that were promoted and uploaded to the staging location on archive.m.o and pushes (copies) them to a final archive.m.o directory. It will also do some final update testing.
Ship - This final phase will make the release live and available on mozilla.org and start serving updates to existing users via our update server Balrog.. See Types of releases below for more.
In most cases, you would trigger each phase individually, one at a time. Note: by design, it is possible to only click a later phase and Shipit is smart enough to backfill tasks from uninitiated earlier phases.
Once you trigger a phase, you can re-click on that phase to monitor the progress of the automation. This monitoring is through the Taskcluster’s Task Monitor UI. You have to keep an eye on these tasks. If any of them fail, escalate it to the Release Engineering team via the proper communications
For more on how to trigger phases of a release in Shipit, including UI screenshots, see Relman’s documentation
Signing off in Balrog¶
If the channel you are releasing is Beta or a mobile product, no further action is required. If you are shipping a ESR or Release release, you also need to sign off in Balrog itself via the Balrog Admin UI
To guard against bad actors and compromised credentials we require that any changes to primary release channels (beta, release, ESR) in Balrog are signed off by at least two people.
After the scheduled change has been created by the “updates” task in the “ship” phase that was triggered in Shipit, and prior to the desired release publish time.
In context of the other rules, eg
Further details and examples can be found on the [[Balrog page|Balrog and Scheduled Changes]]
For more on how to sign off of a Balrog channel/rule change, including UI screenshots, see Relman’s documentation
How these taskcluster tasks are scheduled¶
Shipit uses Taskcluster actions and the “taskgraph” tool to schedule what tasks are needed in each phase. Taskcluster uses special release workers that are under Release Engineering control and locked down to do the actual work. See Source and under the hood below for more details.
Other Release Management Tasks¶
Release MGMT have a number of tasks and coordination items they need to address with each release. See an example of the 76 release checklist
Escalating issues and communications¶
For email, Slack, and Matrix communications with various release stakeholders, see the communications section.
For troubleshooting a release automation issue, contact Release Engineering via above.
For any coordination or product specific issue, contact Release Management via above.
Source and under the hood¶
Firefox is released via the same tooling that’s used to build and test Firefox. We use our Mozilla in-house continuous integration (CI) platform Taskcluster to drive the tasks and workers. The main service in this platform is the Taskcluster Queue. The queue takes requests of tasks and coordinates with a pool of workers to actually conduct the task work. The various scheduling and dependency logic is defined in taskgraph. The workers are trusted, locked down, and owned by Release Engineering. They are scriptworker based and the various implementations live here
We use balrog scriptworkers that interface with Mozilla’s updater service, Balrog