Skip to content

Postmortem: Build Breakage on 2016 11 08

Todd Volkert edited this page Nov 9, 2016 · 2 revisions

Flutter postmortem: Build Breakage on 2016-11-08

Status: final
Owners: chinmay


Description: Travis reported failures on builds
Component: flutter repository
Date/time: 2016-11-07 21:30
Duration: 16h 45m
User impact: Flutter team members were unable to merge new PRs. Users would have been unable to run flutter tests if they upgraded, though we did not receive complaints during the outage.

Timeline (all times in PST/PDT)


A change to package:args is committed (591f9c) that introduces a bug whereby run() no longer returns the value returned by the command.


15:11 The change to package:args is merged into the args repository.


16:27 Dart package:args tag 0.13.6+1 is cut -- and shortly after is pushed to pub <START OF OUTAGE>
21:36 ianh reports that Travis is upset and all PRs are failing


07:52 danrubel reports that Travis is still failing
11:03 chinmaygarde reports he’s facing the same breakage in his pending PR
11:07 Issue is reproduced locally. chinmaygarde, jsimmons and danrubel begin looking for the root cause of the breakage.
13:08 Root cause of outage identified as a new version of package:args that Flutter picked up whereby run() no longer returns the value returned by the command (so we couldn’t get accurate exit codes).
13:19 Flutter PR #6765 sent to pin Flutter to a known good version of package:args
13:42 fb3bf7a identified as root cause of the internal breakage.
14:15 Fix lands. <END OF OUTAGE>

Root causes

A bug was introduced in package:args that was picked up by Flutter.  Flutter was vulnerable to this bug because our external dependencies have open-ended version constraints, so the stability of our codebase is not hermetic. This was an intentional choice; we have experienced this failure mode previously, and have been running on the basis that we are not yet stable enough to deal with the costs of being hermetic.

Action items


Action Item Owner Tracking bug Notes
Pin our external Dart dependencies to specific versions to ensure that our public stability is hermetic. chinmay #6767


Action Item Owner Tracking bug Notes
We should have a continuous monitoring bot that tries to run all our tests ianh #6777






Action Item Owner Tracking bug Notes
Update our package:args dependency to a known good version danrubel PR #6575 Done
Deploy a forward-rolling bot that goes red if our dependencies release a breaking change, and otherwise updates us to the latest versions of everything. ianh #4696

Lessons learned

What worked

  • Once the Flutter team had a clear set of owners for the issue, it was root-caused and resolved quickly.

Where we got lucky

  • The outage did not break users. It likely would have if we had a larger userbase.

What didn't work

  • There were indications of the breakage as early as 2016/11/07 21:30, yet the team didn’t start looking into it in earnest until 2016/11/08 11:00. Once we get to the point where our build is hermetic (so we control our own stability) and we separate production artifacts from development artifacts (e.g., have a release branch), then we should consider providing an SLA, at which time we’d have to create processes around how to maintain that SLA.

Flutter Wiki


Framework repo

Engine repo


Experimental features

Release Notes

Clone this wiki locally
You can’t perform that action at this time.