#51 · Thank you Microsoft — the Microsoft 365 outage (12–13 Nov 2026)
Outlook, Teams calendars and Copilot down for 18+ hours. Servers failing health checks, traffic piling onto the rest, and the cascading failure pattern every backend interview asks about.
Ek line yaad rakho: ek server gira, uska load baaki pe, aur domino shuru.
What happened
- 12 Nov 2026, 16:17 UTC: Exchange Online starts failing (incident EX1550543, also MO1550821). Outlook (web, desktop, mobile), Teams calendar/scheduling, Copilot, Planner and parts of SharePoint are hit; Teams calls and chats mostly keep working.
- Scale: about 130,000 Outlook reports on Downdetector at the peak (AP, via MacRumors); Japan reported impact on government IT systems.
- Duration: still degraded 18+ hours later on 13 Nov; Microsoft extended the email timeout window to 48 hours so delayed mail isn't lost.
- Not a hack: Microsoft says there is no indication of malicious or cyber activity (Windows Latest).
Why (Microsoft's updates so far)
- Servers began failing automated health checks; a process that sets up secure connections was overwhelmed.
- Failed servers were taken out → "servers being unexpectedly unavailable; straining the remaining infrastructure" (Tom's Guide live updates).
- Remaining servers overloaded → more health-check failures → more servers out: a cascading failure.
- A recent network configuration change coincided with the start; Microsoft paused deployments, reverted it, and patched an authentication/routing component, then redistributed traffic worldwide (Forbes, CX Today).
The final root cause will be in Microsoft's post-incident report.
The interview answer: cascading failure
- Load shedding: when overloaded, reject extra work early (429/503) instead of falling over.
- Circuit breakers + retry budgets: stop hammering sick dependencies; exponential backoff with jitter.
- Careful health checks: don't eject too many servers at once (max-ejection percentage); separate liveness from readiness.
- Staged config rollout: canary a config change to 1% of servers/one region; config is code.
- Headroom: run with spare capacity so losing N servers doesn't overload the rest.
Words to say: cascading failure, load shedding, circuit breaker, retry storm, max ejection, staged rollout.
Go deeper
Finished reading?
Mark it done to track your progress in Reels.
Reel Notes
Notes for every @masstdev system design reel — diagrams, interview words and the follow-up questions interviewers ask.
#39 · AWS outage: one empty DNS record
Case study of the October 2025 AWS us-east-1 outage — a race between redundant DNS automation left DynamoDB's endpoint with an empty record. Race conditions and version checks.