Hedged Requests
A tail-latency mitigation technique that launches a duplicate of a slow request on another execution path and accepts the first successful result.
Some services maintain a low mean latency while a small fraction of requests become much slower because of queueing, network variance, or a temporarily slow dependency. Request hedging starts another copy after a delay threshold, often against a different connection, replica, or worker, and keeps the first successful result.
The trade-off is extra load. Hedging can worsen saturation, and duplicate side effects are unsafe unless the operation is idempotent or reliably deduplicated. It therefore belongs in measured tail-latency control rather than as a generic retry substitute. Safe retry design in critical systems covers the adjacent failure boundary.