Q: Should I alert on slow response times, not just full outages?

Q: Should I alert on slow response times, not just full outages?

A: Yes, because most outages announce themselves as creeping slowness first. A response-time alert is your early warning before the site tips over into hard 5xx territory.

But set the threshold off real baselines, not a round number. Pull your normal response-time distribution and alert on a sustained breach of, say, your p95, held for a few minutes, not a single slow sample. A static "alert if over 2 seconds" either screams during every traffic spike or never fires at all.

Two refinements:
— Separate "slow" (warning, business hours, Slack) from "down" (page, anytime).
— Measure time-to-first-byte separately from full load, so you can tell a slow server apart from a slow page full of heavy assets.

Follow-up: require the slowness to persist before alerting, or normal bursty traffic will train you to ignore it.

Got a question? Drop it in the comments.