x/build/cmd/watchflakes: failures posted with misleading timestamps

[bcmills]

In #62095 (comment), watchflakes posted a number of failures in x/ repos with timestamps dating back to June, like:

2023-06-21 11:18 linux-amd64-racecompile mod@baa5c2d0 go@2fcfdb96 go/internal/gccgoimporter [build] (log)

That timestamp is misleading: go@2fcfdb96 was committed 2023-08-17, not 2023-06-21.
(The timestamp is for mod@baa5c2d0, ignoring the go commit time.)

Those kinds of misleading timestamps can easily cause an issue to be mistriaged: a sudden new failure starting two months ago is probably a misconfigured or recently-added builder, whereas a sudden widespread recent failure is almost certainly a bad commit that should be rolled back.

(attn @cherrymui)

[bcmills]

#63730 (comment) is another one with misleading timestamps, which would prevent it from being correctly matched to #63739.

[cherrymui]

Given the current data model with the coordinator, this is rather difficult to do. LUCI will provide us better data. As we're going to switch to LUCI, we will solve it with LUCI.

[bcmills]

I agree that the coordinator makes this difficult. FWIW, fetchlogs uses maintner for this information:
https://cs.opensource.google/go/x/build/+/master:cmd/fetchlogs/fetchlogs.go;l=143-158;drc=a3f1e410a493bb04cbc55bb7c99b81457f6cae66

It's not the most efficient (maintner is kind of a memory hog and takes a short while to load from scratch), but it at least sidesteps the need to make changes to the coordinator.

If watchflakes has enough network access to use maintner, I expect this could have a workaround in place with less than a day's work.