Why Do We Need Consistent Hashing?

When a system grows beyond one machine, it needs a predictable answer to a small but important question: which machine owns this key? A naive answer is to take the hash of a key and use the remainder after division by the number of servers.

const serverIndex = hash(key) % servers.length;

That works while the server list is fixed. The trouble begins when it changes. Adding one server changes the divisor, so a large share of keys now map somewhere else. A cache can suddenly go cold; a storage layer may need to move far more data than necessary.

Put servers and keys on the same ring

Consistent hashing maps both keys and servers into the same circular hash space. A key belongs to the first server found when moving clockwise around the ring.

Server A Server B Server C

If a server leaves, only the keys that would have landed on that server move to its next neighbor. If a server joins, it takes responsibility for a section of the ring rather than reshuffling everything.

Why it helps

• Smaller migrations: a membership change affects a bounded portion of keys.
• Predictable ownership: every client can make the same routing decision from the same membership list.
• Incremental growth: adding capacity does not require a global reset.

In practice, systems often give each physical server several virtual positions on the ring. These virtual nodes smooth out uneven key distributions and make capacity weighting easier.

Consistent hashing does not remove the need to move data. It makes the movement local and manageable.

A small example

Event	What happens to a key owned by Server B?
Server C joins after B	The key stays on B.
Server B leaves	The key moves to the next server clockwise.
A virtual node is added near B	Only keys in that small interval move.

The exact hash function is less important than a stable, shared definition of the ring. Once that contract is in place, clients can route a key such as customer:42 without asking a central directory for every request.

The main idea is simple: move the smallest possible set of keys when the set of servers changes.