Repository navigation
Long delay for new ci.yaml targets to start running automatically #151615
Description
#151433 added the new Linux android_java11_tool_integration_tests ci.yaml target, but this target didn't start getting run automatically in post submit till ~20 hours later.
Things that were notable:
- The
Linux ci_yaml flutter rollerrun for that commit indicated no changes needed to be rolled (despite the commit containing changes to theci.yaml - The bot pool indicated most of the machines were idle, so it wasn't an issue of capacity
- There was a commit to
flutter.googlesource.comthat rolled in the yaml changes (commit here), but that commit strangely had an incorrect commit message:
Roll flutter to b0cbfc6
That commit hash maps to b0cbfc6, which is the commit before my PR - so it shouldn't contain my yaml changes. And yet, looking at the diff for that commit, we can see that it does contain my Java 11 specificci.yamltarget:
+ builders {
+ name: "flutter/try/Linux android_java11_tool_integration_tests"
+ }
Activity
gmackall commentedon Jul 11, 2024
cc @christopherfujino
christopherfujino commentedon Jul 11, 2024
I would add that the first automated (that is, not manually triggered by Gray) build of the new .ci.yaml target (https://ci.chromium.org/ui/p/flutter/builders/staging/Linux%20android_java11_tool_integration_tests/2/infra) started at July 11, 11:53am PDT (-7), which is over 21 hours after infra commit landed https://flutter.googlesource.com/infra/+/cf2f5c1bb5f28969d6223f4cc8c845e25d314909%5E%21/#F0 at July 10, 9:39pm GMT (+0).
Here is the overall builder view for the new target: https://ci.chromium.org/ui/p/flutter/builders/staging/Linux%20android_java11_tool_integration_tests
gmackall commentedon Jul 11, 2024
Looks like for question number 3, the changes got pulled in by the PR before mine:
https://ci.chromium.org/ui/p/flutter/builders/prod/Linux%20ci_yaml%20flutter%20roller/14810/overview
https://logs.chromium.org/logs/flutter/buildbucket/cr-buildbucket/8742764206402536481/+/u/calculate_Change-Id/git_diff/diff__without_hashes_
Perhaps both these commits (mine and the one before) landed at a similar time, and then the
Linux ci_yaml flutter rollertask ran for the commit before mine first, so when it eventually ran for my commit there was nothing left to roll.gmackall commentedon Jul 11, 2024
Capturing offline discussion that the
Linux ci_yaml flutter rolleractually checks outToT(intended to allow builds to run out of order, but not end up reverting and relanding yaml changes a bunch of times), and not the hash that kicks off the task.But if we are going to base the diff that we roll to
flutter.googlesource.comoff ofToTand not off of the commit that we are running theLinux ci_yaml flutter rollerfor, then we should also base the commit message off of thatToThash (and not the hash of the commit that kicked off the task)