Consider a server that spawns multiple initialization threads: one for the database connection pool, another for the cache, a third for loading configuration. The server can't start accepting requests until all of these are ready.

How do you make the main thread wait for all three?

Coordination problems like this require something fundamentally different from locks. Locks protect resources (preventing two threads from modifying the same data). But here, you need to coordinate timing (making threads wait for each other at specific points). That's what barriers and latches are for.

This chapter explores what they are, when to use each, how they work internally, and the subtle bugs that can occur when using them incorrectly.

What Are Barriers and Latches?