Tech in Diagrams

Monitoring vs Observability

Monitoring watches for known problems; observability is being able to investigate ones you did not predict.

Intermediate 4 min Cloud Computing
One incident, from symptom to root cause
  • Request trace
  • Investigation
  • Anomaly
DISTRIBUTED APPLICATIONone request · id 9f2cMETRICSNumbersover timeLOGSEventswith detailTRACESOne requestend to endINSTRUMENTATIONContext comes from the code, not the toolRequest ids, meaningful attributes and propagation between services.KNOWN QUESTIONSDashboardsTHRESHOLD BREACHEDAlert: checkout errorsError rate above 2% for five minutes.Tells you that it is wrongThe rule never encoded why.SYMPTOM → INVESTIGATIONASKING A QUESTION NOBODY CHARTEDWhy are these checkouts failing?METRIC · LATENCY p95A spike at 14:32 narrows the windowSame window used for the logs and the trace below.LOGS · SAME WINDOW, SAME REQUEST IDTRACE · REQUEST 9F2CROOT CAUSEPayments retries a slow third partyCheckout waits, then times out. No dashboard predicted this one.WHICH JOB IS IT?Same telemetry.Different question.
Monitoring watches the signals someone chose in advance and tells you that something is wrong; observability is having enough telemetry and context to work out why when nobody predicted it.

Hover the system, any telemetry type, the dashboards, the alert, the investigation or the comparison rail to isolate related signals; click any component for a short explanation.

Next conceptAI in the CloudCloud AI services expose trained models as APIs, so applications can use them without owning accelerators.