Just use PostgreSQL? Not at 100 million records.

transcript

Where does your data live? Easy. Just use PostgreSQL. Great answer. Until it meets four numbers.

100 million records. 10,000 requests a second, from users on three continents. And 99 point 99 percent uptime. Now just use PostgreSQL isn't an answer.

And your data isn't fair. A tiny fraction of short links gets most of the traffic. Every other link shares what's left. Then one link goes viral. A million hits a second, on a single key.

Your rate limiter loses a Redis node. Over-allow, or under-allow? A datacenter fails over. What does your user see? A user shares a brand new link. A friend clicks, on an async replica. What happens?

A developer asks one thing. PostgreSQL, or MongoDB? A systems engineer asks first. What are the access patterns? What consistency do we actually need? What happens when a node fails? Only then: here's how we store it.

Five days. Split it. Copy it. Throttle it. Survive the hot key. Then a session store. This is the week you stop using databases, and start designing storage.

Full breakdown in the lecture.