Monitoring and Alerting

Monitoring turns system signals into a small number of alerts that are worth waking someone for.

Beginner 4 min DevOps
Signal · rule · alert · response
  • Signal
  • Firing
  • Recovered
PRODUCTION · checkout-apiRunning, and reportingThree instances serving real users, emittingthe same few numbers every few seconds.emitting…AVAILABILITYRequests succeeding99.98%SteadyERROR RATE5xx responses7.4%Above thresholdLATENCY p95Time to respond612 msRising with errorsSATURATIONCPU headroom used61%HealthyALERT RULEA condition, not a numbererror_rate > 5% for 5mThe signal reports. The rule decideswhether that reading deserves a person.THRESHOLD + WINDOWTwo brief spikes discarded · third held 5mALERT STATEOKFIRINGRESOLVEDAn alert is a small state machine: it openswhen the condition holds and closes itselfwhen the signal recovers.HUMAN RESPONSEMitigate, then understandRoll back the deployment that introduced theerrors. Restore service first — the review ofwhy it happened comes afterwards.NOTIFICATION · DELIVEREDPaged the on-call engineerFiring is a state inside the system. Thenotification is how it reaches a human.CRITICAL → phoneWARN → chatACTIONABLE ALERTcheckout-api · errors 7.4%WHATCheckout requests failing for 5 minutesIMPACT≈ 1 in 13 purchases affectedNEXTCompare with the 14:02 deploymentEverything needed to start, in one message.RECOVERY · LOOP CLOSEDRESOLVEDerror_rate 0.4%Errors fell back under the threshold andstayed there, so the rule evaluates falseagain and the alert closes itself — no onehas to remember to tidy up.SEVERITY · ROUTINGWho gets wokenCRITICAL · pages on-callWARNING · team channelINFO · dashboard onlyImpact decides the channel.ANTI-PATTERNAlert fatigueTwenty pages a day and theone that mattered is missed.Alert on symptoms, require awindow, delete rules nobodyhas ever acted on.
Monitoring watches known signals; alerting turns one sustained, meaningful condition into an action for a person — and closes itself when the signal recovers.

Hover the system, any signal, the rule, the threshold and window, the alert state, the notification, severity routing, the incident card, the response, recovery or the fatigue note to isolate related elements; click any component for a short explanation.