Tag
1 note
A tail-latency optimization that issues the same idempotent request to multiple replicas and uses whichever responds first, trading extra compute for dramatically lower P99/P999.