Observability & Operations

Incident Management & Postmortems

An incident is managed in two phases — restore service first, understand root cause second — and a blameless postmortem is how an organization actually learns from an outage instead of just being relieved it is over.
  • Mitigate before you diagnose: rolling back a bad deploy or failing over to a healthy region ends user impact immediately, even before the root cause is known
  • An incident commander owns coordination and communication during an incident so responders can focus on fixing instead of status-updating
  • A severity scale drives response consistently — who gets paged, whether execs are notified, whether a public status page updates — instead of deciding ad hoc each time
  • Blameless postmortems focus on what about the system allowed this, rather than who made the mistake — punishing the person who tripped an incident just teaches everyone to hide the next one
  • A postmortem's real output is action items with owners and dates — a document that identifies causes but assigns no follow-up changes nothing
  • Timeline reconstruction — what was known, when, and what was done about it — is often more valuable than the root cause itself, since it reveals detection and response gaps that recur across unrelated incidents
From detection to a shipped fix
A severity scale example
SeverityImpactWho is pagedComms cadence
SEV1full outage or data losson-call plus incident commander plus leadershipcontinuous, public status page
SEV2major feature degradedon-call plus incident commanderregular internal updates
SEV3minor, contained impacton-callticket, no broad comms
Sources
  • Site Reliability Engineering: How Google Runs Production SystemsCh. 14-15 — Managing Incidents; Postmortem Culture