Originally composed as IRC messages by the venerable rnewson
.
In Apache CouchDB 2.0, there's no such notion of committed. It's AP not CP.
What happens is a write is applied to N nodes (default is 3) and a 201 response is returned when the first W nodes (default is 2) respond with success. A 202 response is sent if < W writes (but > 0) were made. Every node, on receiving an update, does an automatic replication to the others (using an internal, faster, system than _replicate).
Documents are read from N nodes that hold a copy, and a 200 response is sent (and the document) when the first R nodes respond with the same revision.