Couchbase cluster across AWS availability zones?

The documentation says that "Couchbase Server is not designed to have clusters or clients that span data centers. It requires that intra-cluster (server to server) latency be very low, as well as between the cluster and Couchbase clients. If you require the ability to be in one or more data centers, see Cross Datacenter Replication (XDCR).”

Amazon’s availability zones are separate data centers (in order to avoid correlation of failures), but the availability zones in a region are located close together in order to reduce latency. I get round-trip (ping) times of 0.3 ms between servers in the same availability zone and 0.6–1.6 ms between availability zones. Is this little enough that Couchbase will be happy with the intra-cluster latency?

What are the implication of network partitioning, so one AZ (and the Couchbase server(s) there) becomes separated from the other, but still available to some clients?

Stretching Couchbase Server across AZs is common best practice. We are perfectly happy to be stretched across AZs in the single region. We’ll clarify the documentation. Make sure also to look at Rack and Zone awareness capability to help Couchbase Server distribute replicas with AZ awareness as well when deploying.

submitted this correction for clarity:

Thank you very much!