A network partition happens when parts of a distributed system can no longer talk to each other reliably.

The machines may still be running. The processes may still be healthy. From the outside, both sides may even look like they are serving traffic normally.

What the system has lost is coordination.

That is what makes partitions dangerous. A crashed node stops making changes. A partitioned node may keep making changes while nodes on the other side make different changes at the same time. Neither side knows what the other side is doing until the network heals.

When communication comes back, the system has to answer a hard question: what do we do with two different versions of history?

This chapter explains why partitions happen, how they differ from ordinary crashes, and what choices a system has when one side cannot reach the other.