Alerting on consecutive failures vs. a failure-rate window — which beats flapping?

Alerting on consecutive failures vs. a failure-rate window — which beats flapping?

Q: My monitor pages me, recovers, pages again. How do I stop the noise?

A: That's flapping, and the fix depends on the pattern. "Alert after N consecutive failures" (say 3 in a row) is simple and great for clean, total outages — but a service that fails every other check never hits 3 in a row, so it slips through.

A failure-rate window catches that: "alert if 4 of the last 6 checks failed." It tolerates the odd blip but flags genuine instability that consecutive-only logic misses.

Use consecutive-failure logic as your default, and switch to a rate window for any service that flaps intermittently rather than dies cleanly. Pair either one with a recovery delay — require 2–3 healthy checks before declaring "resolved" — so you're not paged on every brief recovery during a shaky period.

Got a question? Drop it in the comments.