api: do not log controller-runtime messages after context cancel

[milas]

This is mostly annoying in tests when trying to reproduce an issue
and running with something like -count 100. With the async nature
of the reconcilers, it's pretty inevitable that some things will
still be in-flight when the tests are done (and so cancel the context
to clean up resources), and a bunch of irrelevant errors that are
really just wrapped "context canceled" get logged out.

[milas]

FYI this isn't perfect, you still can get stuff like:

E1122 16:27:26.765708   72779 wrap.go:54] timeout or abort while handling: method=GET URI="/apis/tilt.dev/v1alpha1/uiresources/%28Tiltfile%29" audit-ID="201a6f2b-df7d-405a-8ba3-2c86feb57f01"
E1122 16:27:26.765806   72779 timeout.go:135] post-timeout activity - time-elapsed: 11.253µs, GET "/apis/tilt.dev/v1alpha1/uiresources/(Tiltfile)" result: <nil>

I'm guessing these are doing something sketchy like writing directly to os.Stderr 😬

We already have some other string patterns we ignore, I held off on adding these for the moment, but I think timeout or abort while handling and post-timeout activity are probably both phrases we can add to the filter.

[landism]

huzzah! this was really spammy!

you still can get stuff like

fwiw, it looks like these are just writing directly to klog w/ severity 2, which doesn't seem toooo sketchy and also wouldn't go through the filter, I assume?

tilt/vendor/k8s.io/apiserver/pkg/server/filters/wrap.go

Line 54 in 948bda7

runtime.HandleError(fmt.Errorf("timeout or abort while handling: method=%v URI=%q audit-ID=%q", req.Method, req.RequestURI, request.GetAuditIDTruncated(req)))

tilt/vendor/k8s.io/klog/v2/klog.go

Line 1456 in 948bda7

logging.printDepth(errorLog, logging.logr, logging.filter, depth, args...)

[milas]

Interesting - I was still seeing them sporadically in the logs when testing. I guess something else funky is going on there 🤷