Cards/Distributed DataSystem Design · Day 30Aug 30, 2026

Why Copies Disagree

Leader and follower replication — synchronous waiting for confirmation against asynchronous replying immediately, and a timeline showing followers landing the same write at different moments

One database is a single point of failure, so you keep copies, and now you have a harder problem.

The shape is almost always the same: a leader takes every write and ships it to followers, which serve reads. The only real question is when the leader says yes.

Synchronous means waiting for a follower to confirm before replying. Safe, slow, and if that follower stalls, writes stop everywhere.

Asynchronous means replying immediately and shipping afterwards. Fast, always available, and if the leader dies inside that window the write is simply gone.

Most systems run mostly async, because one slow follower should not be able to halt the business. Which leaves the copies behind — continuously, as a standing condition rather than an occasional hiccup.

A write accepted at T0 reaches one follower at T1, another at T2, a third at T3. Read from any of them and you get a different answer, all of them legitimate, each true at a slightly different moment.

The moment you make a copy, you stop having one truth and start having a timeline.