Real-World Dissections: Famous Architectures & Outages, Explained
Twelve primary-source case studies (GitHub, Discord, Uber, Hotstar, Slack, Cloudflare, AWS, Meta, Figma, Notion), each told as context, timeline, root cause, what changed, lessons, and how to cite it in an interview
About 150 minutes · 18 exam questions · 12 interview questions · 2 code labs
What you will learn
- GitHub, 21 Oct 2018 — 43 seconds of partition, 24 hours of degradation
- GitHub gh-ost — why ALTER TABLE on a busy MySQL primary needed a new tool
- Discord — trillions of messages: Cassandra → ScyllaDB, with Rust data services
- Uber, 2016 — "Why we switched from Postgres to MySQL", and the counter-arguments
- Hotstar — 25 million concurrent viewers: pre-warm, ladders and panic mode
- Slack, 4 Jan 2021 — the first Monday back, cold caches and a Transit Gateway
- Cloudflare, 2 Jul 2019 — one regex, every CPU, 27 minutes
- Cloudflare, 18 Nov 2025 — a generated config file that doubled in size
- AWS us-east-1, 19–20 Oct 2025 — the DNS race that emptied DynamoDB's endpoint
- Meta, 4 Oct 2021 — the backbone command and the DNS servers that hid themselves
- Figma, 2020–2024 — horizontally sharding RDS Postgres without leaving Postgres
- Notion, 2021 — 480 logical shards on 32 Postgres hosts, keyed by workspace
- The pattern map — five failure shapes, and how to tell any case study in 90 seconds
Previous: Incidents IV: Correctness, Regions & Operability · Next: Architecture Evolution Stories: Airbnb, Netflix & Shopify