During a write request by a client, the coordinator node finds the replicas for the partition key based on the partitioner being used. It then sends the write request to all the nodes that are owners of that partition key range. Coordinator nodes return write success to the client as soon as many number of nodes return write success based on the consistency requirement of the write operation. Let's assume that three nodes are replica nodes for a write operation and the write operation is being done at QUORUM consistency; whenever two out of three nodes return success for the write operation, the coordinator node acknowledges a write operation success to the client. We'll learn more about consistency in the next sections.

On a node, a write request is first written to a file called CommitLog, and then in a memory data structure called a Memtable. Whenever Memtables are full, they are flushed to disk in files called SSTable. A replica node acknowledges a write operation success...