Thanks for the fix 💯

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @adityamaru, @andreimatei, and @stevendanna)

pkg/util/tracing/crdbspan.go line 547 at r1 (raw file):

		// We don't have space for this recording. Let's collect just the structured
		// records.
		for ci := range childRecording {

I think a better fix is to get rid of the assertion in the structured case below, and make that code deal with multiple spans too. That's how it was before #81079, and I'm not sure if there was a specific reason why it changed.

I think it's important for all this code to be fairly permissive in what recording it accepts. For example, as written, I think this code would crash if an RPC is made within a RecordingStructured span, and then the remote child is dynamically changed via \tracez to RecordingVerbose: a verbose, multi-span recording will come back and the client span will choke on it.

Please add the following test with this occassion:

// Test that a RecordingStructured parent does not panic when asked to ingest a
// remote verbose recording. Ingesting a recording of different type is unusual,
// since children are created with the parent's recording mode, but it can
// happen if the child's recording mode was changed dynamically.
func TestRemoteSpanWithDifferentRecordingMode(t *testing.T) {
	tr := NewTracer()
	s1 := tr.StartSpan("p", WithRecording(tracingpb.RecordingStructured))
	s2 := tr.StartSpan("c", WithRemoteParentFromSpanMeta(s1.Meta()), WithRecording(tracingpb.RecordingVerbose))
	s3 := tr.StartSpan("cc", WithParent(s2), WithRecording(tracingpb.RecordingVerbose))
	s3.Finish()
	r := s2.FinishAndGetConfiguredRecording()
	require.NotPanic(s1.ImportRemoteRecording(r))
	s1.Finish()
}

pkg/util/tracing/span_test.go line 408 at r1 (raw file):

}

// TestRecordingFinishWithMaxSpans is a regression test to ensure that a verbose

let's rephrase this test as Aditya was suggesting, and then I don't have to nitpick about this comment.
Call it TestRecordingDowngradesToStructuredIfTooBig and check that you end up with a recording with just one or two spans or whatever, and all the structured events.

TFTR!

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @adityamaru and @andreimatei)

pkg/util/tracing/crdbspan.go line 547 at r1 (raw file):

Done.

pkg/util/tracing/span_test.go line 408 at r1 (raw file):

Done. We don't get all of the structured events because of the size of the ring buffer, but I've checked that we get as many as possible given the size of the buffer.

but let's wait for Aditya to seewhat he says about that assertion. I think at some point he was relying on it, but maybe that changed.

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @adityamaru, @andreimatei, and @stevendanna)

pkg/util/tracing/span_test.go line 460 at r2 (raw file):

	s3.Finish()
	r := s2.FinishAndGetConfiguredRecording()
	require.NotPanics(t, func() { s1.ImportRemoteRecording(r) })

please also test that s1's recording only has one span in it