Launch readiness review — Diwali sale in 10 days; marketing expects 8–10× normal traffic for 6 hours
Is the Diwali sale ready?
You're running the production readiness review with the ShopLite team.
Briefing
Last year's sale peaked at 6× traffic and caused a 40-minute checkout outage when the database ran out of connections. This year's forecast is higher. You have ten days.
Your shift
Decide before you look. Each choice is locked once made; the next decision unlocks after it. There's no perfect path in real incidents either, only better and worse judgement under uncertainty.
The team says 'autoscaling will handle it'.
What's your first priority?
Debrief
Readiness is evidence, not optimism: load-test the real bottlenecks, fix what breaks first, prepare degradation switches for what you can't predict, freeze changes around the event, and staff an incident team for the peak.
The postmortem
Blameless postmortem · draft
Summary. Diwali sale served a 9.2× peak with checkout p99 of 640 ms and no SLO breach.
Impact. No customer impact; the recommendations widget was switched off for 25 minutes at peak as a precaution.
Timeline
- Day -10Load test finds DB connection exhaustion at 4×
- Day -7RDS Proxy added; 10× load test passes
- Day -2Change freeze begins
- SalePeak 9.2×; one degradation switch used
Root cause. N/A
Contributing factors
- Last year's outage (connection exhaustion) directly informed this year's load test plan.
Action items
- preventMake the 10× load test part of the quarterly readiness routinesre
- mitigateKeep degradation switches permanent and tested monthlyshoplite
The practice behind it
Production readiness reviews
A structured checklist before launches and big events: SLOs and alerts defined, dashboards and runbooks ready, capacity tested beyond expected peak, dependencies and their limits known, rollback and degradation plans rehearsed, on-call staffed, and a change policy agreed. It turns 'I think we're ready' into specific, verifiable answers.
Your turn
Name three things that commonly break first under a traffic spike, before CPU does.
Interview questions
How do you prepare a system for a known traffic spike?