The mechanics of consistent hashing

With consistent hashing, the buckets are arranged in a ring with a predefined range; the exact range depends on the partitioner being used. Keys are then hashed to produce a value that lies somewhere along the ring. Nodes are assigned a range, which is computed as follows:

For a five-node cluster, a ring with evenly distributed token ranges would look like the following diagram, presuming the default Murmur3Partitioner is used:

In the preceding diagram, the primary replica for each key is assigned to a node based on its hashed value. Each node is responsible for the region of the ring between itself (inclusive) and its predecessor (exclusive).

This diagram represents data ranges (the letters) and the nodes (the numbers) that own these ranges. It might also be helpful to visualize this in a table form, which might be more familiar to those who have used the nodetool ring command to view Cassandra's topology.

When Cassandra receives a key for either a read or a write, the hash function is applied to the key to determine where it lies in the range. Since all nodes in the cluster are aware of the other nodes' ranges, any node can handle a request for any other node's range. The node receiving the request is called the coordinator, and any node can act in this role. If a key does not belong to the coordinator's range, it forwards the request to replicas in the correct range.

Following the previous example, we can now examine how our names might map to a hash, using the Murmur3 hash algorithm. Once the values are computed, they can be matched to the range of one of the nodes in the cluster, as follows:

The placement of these keys might be easier to understand by visualizing their position in the ring.

Now that you understand the basics of consistent hashing, let's turn our focus to the mechanism by which Cassandra assigns data ranges.