With RF=3 will data be replicated 3 times in each rack?

Question

We have 9 physical servers, each server has 4 physical disks, which means we have a total of 36 physical disks, each one related to one Cassandra node.

We would like to have a replication of 3 across the full cluster, in a way that would allow us to lose one server (4 physical disks) without losing data.

We were thinking about making three different racks, each rack with 3 of our physical servers. My understanding is that, if I have a keyspace with RF=3, a set of data should be replicated 3 times across the cluster.

My question is, will that data be replicated 3 times inside each rack, or will it be replicated once in every rack (having 3 racks)? what should I do/how should I configure my cluster and my keyspace to have the data replicated exactly one time for each one of our three racks?

Answer 1

As I've explained in your other question about how replication works, Cassandra will pick replicas based on the topology if the keyspace is configured with NetworkTopologyStrategy such that each of the three racks in the DC will have one copy of the data.

The idea behind this is that if any of the nodes in a Cassandra rack is unavailable for whatever reason, 2 other copies (replicas) in the other 2 racks are available to serve requests for consistencies of LOCAL_ONE or LOCAL_QUORUM.

To explicitly answer the second part of your question, Cassandra will NOT store 3 copies of the data in each rack -- just ONE copy per rack. Cheers!