Design the Whole Thing (the interview, assembled)

The capstone board of the System Design track. Every earlier board taught one box. This one takes a blank prompt, “design a URL shortener”, and shows which boxes the numbers pay for, and which ones the numbers refuse.

Before you start. You need the five things the earlier lessons built, and one clause each is enough to bring them back: a load balancer (one public address in front of many identical servers), a cache (a small fast copy of the answers people ask for most), a read replica (a second database that only answers reads), a shard (one database split across machines by key), and a log (an append-only record of things that happened). You also need to be able to multiply and divide numbers like 100 million and 86,400 in your head, badly, to one digit.

The idea in 30 seconds

An open-ended design question is six questions in a fixed order, not an invitation to invent something.
Nothing goes on the whiteboard until a number pays for it, and the most valuable numbers are the ones that take a box off.
A day is about a hundred thousand seconds. That one conversion turns any “per month” figure into QPS in your head.
A hundred million new links a month is 40 writes a second and 3 TB over five years, which is one machine, which is why a shortener does not need sharding.
The cache hit rate, not the read rate, decides whether you need replicas. Same traffic, different architecture.
A counter never collides; a truncated hash in the same space collides about five million times. Both fit, only one is free.

Read the story · Skip to the numbers