Course contentsShow
← Software Engineering
Lesson 1181 of 1,93335. Monitoring and AlertingFree lesson

When to Conduct a Postmortem

Identifying which incidents warrant formal postmortems versus lighter-weight reviews or no review at all.

When to Conduct a Postmortem

What you'll learn: You'll learn the criteria for deciding which incidents deserve a full blameless postmortem versus lighter reviews or no formal analysis.

Not Every Incident Needs a Postmortem

Just as you wouldn't hold a formal investigation every time someone stubs their toe, not every system hiccup warrants a full postmortem. The goal is to learn from significant events without creating administrative overhead that dilutes focus or causes "postmortem fatigue."

Clear Triggers for a Postmortem

You should conduct a postmortem when:

  • SLO violations occur — Any breach of your error budget or service level objectives signals something meaningful went wrong
  • Customer impact was significant — Even brief outages affecting many users or critical users deserve analysis
  • On-call engineers escalated — If the incident required multiple people or reached senior engineers, it was serious enough
  • Novel or surprising failures — New failure modes are learning opportunities, regardless of customer impact
  • Near-misses with potential — "We got lucky" situations where disaster was narrowly averted

What Doesn't Need a Full Postmortem

Consider lighter-weight reviews or no review for:

  • Incidents caught by canary deployments before production impact
  • Known issues with documented workarounds already executed successfully
  • Minor alerts that auto-resolved within seconds
  • Issues completely contained within a single service with no propagation risk

Think of it like triage: serious injuries get full emergency room treatment, minor cuts get a bandage and observation.

The "Would We Learn Something?" Test

When in doubt, ask: "If we spent two hours analyzing this, would we discover actionable improvements?" If yes, do the postmortem. If the answer is "we already know exactly what happened and how to prevent it," document briefly and move on.

Key Takeaway: Conduct formal postmortems for incidents that violate SLOs, cause significant customer impact, involve escalation, reveal new failure modes, or represent near-misses—but avoid postmortem fatigue by skipping analysis of trivial, well-understood, or fully-contained minor issues.