x/build/cmd/watchflakes: currently in a CrashLoopBackOff state

[cagedmantis]

Watchflakes is continuously crashing.

watchflakes-deployment-68796dfdb7-th9zv 0/1 CrashLoopBackOff 986 (3m28s ago) 15d

The last logged item is:
watchflakes: 2024/10/29 17:21:46 repo mismatch: tools go https://ci.chromium.org/b/8733044805650708721

@golang/release

[gabyhelp]

Related Issues and Documentation

• build: golang.org/x/telemetry/crashmonitor failures #67448 (closed)
• x/build/cmd/watchflakes: fails to run after a build output property field gitilesCommit → gitiles_commit rename #69609 (closed)
• build: golang.org/x/telemetry/crashmonitor failures #68214 (closed)
• build: golang.org/x/telemetry/internal/crashmonitor failures #67449 (closed)
• build: golang.org/x/telemetry/internal/crashmonitor failures #68215 (closed)
• build: golang.org/x/tools/go/packages failures #67899 (closed)
• build: runtime failures #67560 (closed)
• build: runtime failures #68068 (closed)
• build: golang.org/x/tools/gopls/internal/cmd failures #68193 (closed)
• build: golang.org/x/tools/cmd/deadcode failures #67417

_{(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)}

[dmitshur]

The input of the LUCI build 8733044805650708721 is https://go.googlesource.com/tools/+/889abb17e125bb0f5d8de61bb80ef15fbe2a130d, which is indeed a revision in x/tools, yet the builder "gotip-..." is for the main Go repo, not x/tools.

It's also the very first build of that builder – maybe it was triggered manually and the "tools" part of revision was intended to be "go". After all x/tools doesn't have a commit with that ID, but the main Go repo does. The second build's revision is in the right repo. CC @mknyszek.

In any case, it seems watchflakes can handle this case of a strange build (mismatch of project between builder definition and input revision) by skipping over the build. It's probably not possible to do anything more in the general case.

[mknyszek]

Oh no! Yes, it's totally possible I triggered this build manually and erroneously, as I did trigger manual builds on that date.

Perhaps such builds should error out sooner. They will work, but they don't make much sense.

[cagedmantis]

I deployed a version which ignores the mismatch case. I'm only seeing two CLs where this is an issue:

watchflakes: 2024/10/29 19:10:22 repo mismatch: tools go https://ci.chromium.org/b/8733044805650708721
watchflakes: 2024/10/29 19:10:22 repo mismatch: tools go https://ci.chromium.org/b/8733044811246345153

[gopherbot]

Change https://go.dev/cl/623995 mentions this issue: cmd/watchflakes: do not exit in the case of a mismatch