Consistent Hashing Ring & Virtual Nodes

Mental Model

Consistent Hashing is a key distribution scheme where both servers and keys hash to the same circular range $[0, 2^{32}-1]$. Walking clockwise from a key’s hash position on the ring identifies the first server encountered, which owns that key. This decouples data layout from server counts and prevents global re-sharding during cluster scaling.

Components and Why They Exist

Client

The client application initiates read and write operations, hashes key values, and locates target cache servers. Request particles leave the Client card, pass through the hash calculation, and route toward the matching clockwise server on the ring.

Hash Function

The hash function deterministically maps both server identifiers and data keys into numeric coordinates within a shared circular keyspace. String keys and node IDs enter the hash function card and convert into degree positions along the ring circumference.

Hash Ring

The hash ring represents the circular $[0, 2^{32}-1]$ keyspace and defines the clockwise ownership boundaries between nodes. The 360-degree ring visualization displays server markers and key tokens, tracing the clockwise arc from each key to its owning server.

Cache Servers (Physical Nodes)

The physical cache servers store key-value partitions in memory and serve client read and write queries. Each physical server card at the bottom of the canvas receives routed request payloads for every token range mapped to its physical or virtual nodes.

Virtual Nodes (Vnodes)

Virtual nodes prevent hotspots and uneven key distribution by placing multiple logical tokens on the ring for each physical server. Interleaved colored arcs around the ring show how vnodes slice the keyspace into smaller, evenly balanced segments across all physical hosts.

Edge Cases and Misconceptions

Virtual Node Density

A common misconception is that a few virtual nodes are enough. In production, systems like Cassandra or Memcached often use 32 to 256 vnodes per physical node to achieve tight load balancing tolerances.

Cascading Failures on Crash

When a node crashes in a simple ring without vnodes, 100% of its load shifts directly to its single clockwise neighbor. If that neighbor gets overwhelmed and crashes, a cascading failure can wipe out the entire ring. Virtual nodes distribute the failed node’s keys across all remaining servers, preventing localized overloads.