Cardinality that kills the metrics bill

That is the mechanism: every distinct label value creates a series. User ids, request ids and full URLs never belong in labels. Put them in logs and traces, where high cardinality is the point rather than the problem.

Tail sampling: buffer the spans and decide after the outcome is known, keeping everything that errored or exceeded a latency threshold plus a small random slice of the rest. Costs more memory at the collector and is almost always worth it.