Twitter (X) - the case study

The whole set on one board; the map and the seven problems it’s built from are linked below.

Boards

Before you start. All you need is the shape of a social feed: you post something, and the people who follow you expect to see it soon after. This lesson is about what happens to that simple idea once “the people who follow you” ranges from three hundred to a hundred million.

The idea in 30 seconds

Reading a feed happens far more often than posting, so doing the expensive work at post time (push) beats doing it at read time (pull) almost everywhere.
Push breaks the moment an account’s followers number in the tens or hundreds of millions, because all of that work lands in a single instant.
The fix isn’t picking one for the whole platform - it’s splitting accounts in two, and pulling only the few huge ones in at read time, merged with everyone else’s precomputed list.
A timeline is a list of markers, not posts. Hydration fetches the real posts and their current counts at the last moment, which is what keeps a like count correct everywhere without rewriting millions of copies.
An id with the moment it was made stamped inside it can be handed out by any machine, with no coordination, and still sorts roughly by time.

Read the story · Skip to the numbers