A cascading failure took down checkout for 2 hours. Lead the technical post-incident review.

Short answer

Run a blameless review: build a precise timeline, identify the trigger and the contributing factors (missing timeouts, retry storms, shared pools, alerting gaps), assess detection and recovery, and agree on prioritized actions with owners (resilience patterns, load shedding, bulkheads, better SLO alerting, runbooks, chaos tests), then track them to completion.

Detailed answer

Structure: summary and impact (orders lost, duration), timeline, trigger (e.g. slow recommendation service), why it cascaded (no timeouts, shared thread pool, retries ×3 at two layers), detection (alerts fired 20 minutes late on CPU, not on checkout SLO), response (manual scaling made it worse), what went well. Actions: timeouts and circuit breakers on non-critical dependencies, bulkheads, retry budgets, SLO burn-rate alerting, runbook for degraded mode, chaos experiment to verify. Track and report closure.

What the interviewer expects

Systems thinking about contributing factors and concrete engineering actions, blameless culture.

Common mistakes to avoid

Stopping at the trigger ("the recommendation service was slow") and blaming an individual.