Capstone: Redis, Everything

The finale of the build, where every earlier wall becomes one assembled desk, alongside the project lesson Build Your Own Redis - From Scratch.

Before you start. This series builds a small copy of Redis, a program that keeps everything it stores in memory instead of on disk, so that asking it something and hearing back feel like the same instant. Seven earlier stages already built everything this toy has: a connection you read and write like a file, an agreed way of marking where one command ends and the next begins, a table that finds any key in one step instead of searching for it, one worker that answers every client without ever standing still waiting on one of them, a way for a key to carry its own expiry moment, a way to clear out old keys when room runs short, and a log written to disk so a crash can’t erase what a client was already told was saved. This lesson adds nothing new to that list. It sends one ordinary command across all seven pieces, in the order it actually happens.

The idea in 30 seconds

Redis keeps everything in memory, so even a small command like SET ... EX 60 has to cross a connection, a table, a clock and a log before it’s really done.
A connection is a socket: a two-way pipe with no start or end marks of its own, so the first job is always agreeing where one command stops.
One worker answers every client by asking which ones have something to say right now, instead of waiting at each one in turn.
A key finds its own spot in the table by turning itself into a number; nothing is ever searched for.
An expiry is stored as an exact future moment, a deadline, not a duration, because a duration only means something at the instant it was said.
The change reaches the log before the client is told it worked, so a crash can’t erase something the client already believes is true.

Read the story · Skip to the numbers