...

Redis is a high-query-performance database. The main use case for it is as a caching layer. Partly due to this fact, data durability isn't of tremendous importance, and the main operating mode is single-node. However, a clustered version was released in 2015. Reading the documentation, it sounds like the intent is more to provide a simple mechanism for transcending simple-node mode than to provide a true distributed database with the kinds of guarantees one would expect (like consistency). Accordingly, the data distribution and primary/replica handling are quite simplistic. This is not meant to be disparaging. Redis Cluster is designed to serve specific use cases well at the expense of other use cases.

In Redis Cluster, keys are hashed into one of 16384 buckets (called slots). The hashing algorithm does not need to be consistent because the slots are fixed in number. Nodes in the cluster have unique and permanent IDs. Cluster integrity is maintained via a gossip protocol, and each node ultimately maintains a connection to every other node. Nodes don't proxy queries to non-local keys. Instead they respond with the hash slot that contains the key, along with the node that currently holds that hash slot (this is similar to Streams). Every node in the cluster is assigned some subset of the slots, and the slots can be manually assigned, unassigned, and moved to, from, and between nodes.

Redis Cluster doesn't have primary and replicas for each slot (the way that Streams has primary and standbys for each task). Rather, it has primary and replicas for each node . Thus, if there are two nodes, A and B with replicas Ar and Br, the node Ar is responsible for replicating the exact same set of slots that are assigned to A, and likewise with B and Br. This simplifies the assignment considerations somewhat, as a given node only  needs to host active (primary) copies or  passive (replica) copies. The replicas can serve queries if the query specifies that it accepts stale data, but most likely this arrangement results in the replica nodes being underutilized. So, my quick judgement is that the simplicity of this arrangement is enviable, but we probably don't want to follow suit.

...

Old Version of KIP-441

...