|RFC-0214: Fuchsia churn policy
Enumerate rights and responsibilities to reduce engineering time spent on change.
|Date submitted (year-month-day)
|Date reviewed (year-month-day)
Update Fuchsia's governance model to include platform churn policies by adding requirements to report the cost of changes, proactively notify impacted teams, and limit the amount of work placed on impacted teams.
Fuchsia is now a large project with several independent teams working hard to meet the goals of various customers. At the same time, platforms like Fuchsia must evolve in ways that require intermittent effort from many contributors across the codebase, informally referred to as "churn".
- Align Fuchsia engineering practices with the project's stated support and stability goals.
- Accurately estimate the time spent on migrations and other externally imposed changes.
- Clarify the boundary between RFC designs and FPS staffing decisions.
- Provide a collection of migration strategies for teams to consider.
- Change the strategy for large scale changes that have already begun.
- Decide what changes should be made.
- Mandate a specific migration strategy for all changes.
- Reduce the rate of changes.
- Set policy for changes that require no effort from client teams.
This RFC was initially written as a Google-internal document shared widely within the Fuchsia team, including members of the Fuchsia Eng Council.
This proposal is a change to Fuchsia's governance model, laid out in the "Impacted teams" and "Initiating teams" subheadings. If this RFC is accepted, it will be implemented by updating the FEC charter, adding a governance policy, and amending the new contributor guide.
Informally, a change is considered "Fuchsia-wide" if it requires development effort or workflow changes from other contributors. This includes changes that impact the ABI, any public portion of the SDK, the contents of a product, or generated code.
All Fuchsia-wide changes will incur an engineering cost for many Fuchsia contributors. This policy centralizes those costs so that they can be minimized. For the purposes of this policy, an "impacted team" is any team that must either approve or make changes to their own code, workflows, or docs in order to accommodate churn.
It is the responsibility of the contributor or team that initiated the change to resolve any breakages. Most of the time, this will mean taking a "revert first, diagnose second" approach.
If your team is impacted by an approved change:
Respond to incoming CL reviews or other changes within two business days.
Support the change author in arriving at a satisfactory conclusion, such as by approving their work, responding to surveys, clearly rejecting proposed changes, requesting specific modifications to the change, and/or answering questions from the change author about the subject matter.
If more than 10% of your time is spent responding to churn, you may flag this issue with email@example.com
If the change is in a style guide (e.g. lints, compiler warnings, etc.), you decide how quickly to resolve new style violations.
Without the churn policy, there are no formal requirements or expectations for changes that create churn. This section adds responsibilities for the authors of such changes.
If your team is initiating a change and will be doing 100% of the work:
- Send mail to the FEC explaining the minimal impact to other teams.
- Send mail to firstname.lastname@example.org to notify contributors of the migration.
- Proceed with the migration.
If your team initiates a change that will require substantial effort from others, including changes initiated by an RFC:
- Create a plan that demonstrates to the FEC that your team will expend at least 80% of the manual effort that is not addressed by automation. Plans must include a list of the impacted teams and the estimated cost of the change.
- After the FEC approves your plan, notify impacted teams. They must be able to schedule the work using quarterly planning over at least two quarters, so notify teams more than a week before the quarter begins.
- Proceed with the migration.
This policy does not impact any active migrations.
To assist change authors, example migration plans will be provided on fuchsia.dev.
Drawbacks, alternatives, and unknowns
The main drawback of this proposal is the required involvement of the FEC at the beginning of the process. It is important that change authors receive timely responses, both from the FEC and from impacted contributors.
One alternative policy would require impacted contributors to approve changes. This has the benefit of always allowing change authors to make progress. This alternative was dismissed because it removes an opportunity for impacted contributors to give feedback about the value and harms of a proposed change.
Another alternative is a more rigid approach, where all migrations must be soft transitions using available versioning mechanisms and a slower pace to spread the effort across all Fuchsia contributors. This differs from the proposed policy by setting no engagement expectations for impacted teams. This alternative was dismissed because it sets no bounds on the review timeline, making it difficult for initiating teams to plan and make timely progress.
Prior art and references
Is there any background material that might be helpful when reading this proposal? For instance, do other operating systems address the same problem this proposal addresses?