Add RFC to disable Chromium stability checks

rfcs/disable_chrome_stability.md

@@ -0,0 +1,66 @@
+# RFC 131: Disable wpt-chrome-dev-stability runs
+
+## Summary
+
+The wpt-chrome-dev-stability job on WPT fails quite frequently and
+it's impacting developer productivity. The job should be changed to
+non-blocking, while keeping a record of potential flakiness
+introductions that could be useful in further investigations.
+
+## Details
+
+Stability jobs are intended to help us ensure that new and modified
+tests aren't flaky. In particular it's designed to ensure that the
+test author is aware when they have written a test that is flaky in a
+different browser. This is because authors writing a test in a
+specific browser engine might be unaware that they are depending on
+details of that implementation that make the test unreliable in other
+engines. Leaving this to other developers to fix is expensive, because
+they have to spend time becoming familiar with details of the test
+which are already understood by the original author. In addition, when
+there are many unstable web-platfrom-tests, it undermines confidence
+in the overall suite and, where such instability is automatically
+handled in CI systems (e.g. by marking a test as [PASS, FAIL] in
+metadata), reduces the ability of the tests to catch regressions.
+
+Currently, stability checks for the corresponding browser are skipped
+when [merging export PRs from their own
+repositories](https://github.com/web-platform-tests/wpt/issues/29737)
+(i.e. Chromium exports don't run Chrome stability checks, and
+likewise for Firefox). This is because we assume that any real
+stability issue is likely to have been handled in the source
+repository. It's common for such changesets to include code changes
+to the browser itself that fix intermittents, and in the source
+repository the tests will run together with the corresponding browser
+changes. However on GitHub we are using the latest development
+release of the browser, which is unlikely to contain code fixes at
+the time of export. This makes the stability checks on these PRs
+highly prone to misleading failures.
+
+The main problem with the stability checks as currently implemented is
+that although the test author is in the best place to understand the
+behaviour of the test, they may not understand how to reproduce the
+failure in another browser, or may be unable to fix the problem
+(e.g. if it's a browser bug rather than a test bug).
+
+Faced with these tradeoffs, the Chromium developers believe the
+project is better served by allowing tests which are unstable in
+Chrome to land in web-platform-tests and using the tooling they have
+available in the Chromium CI system to investigate the
+flakiness. Therefore the job will be marked as non-blocking, so we are
+able to see where it fails and assess whether the correct tradeoff has
+been made.
+
+Marking a job as non-blocking will require changes in the taskcluster
+configuration so that the job does not block the sink task.
+
+Disabling the job entirely would also reduce the CI load. This can be
+considered once the impact of the change is well understood.
+
+## Risks
+
+New intermittent failures specific to Chromium could be introduced
+upstream and only noticed when a WPT import to Chromium CI
+happens. Chromium developers believe the tradeoff is worthwhile, as
+the Chromium project has robust tooling to investigate and handle such
+cases.