Raft
A consensus algorithm that keeps a replicated log identical on a group of servers. An elected leader appends each entry and commits it once a majority has stored it, so the group keeps working when a minority fails and never commits two different values. Ongaro and Ousterhout designed it in 2014 to be easier to understand than Paxos; etcd and CockroachDB use it.
Data
In Quorum#
Distributed SQL (Consensus databases research, which needs Synchronous replication) runs SQL on Raft. Every write waits for the round trip to a majority of the regions it spreads over, while a read asks the leaseholder and costs a quarter of a write. It serves 6,000 transactions per second at size 1 and keeps one leader through a Network partition, so that Incident is a quarter as likely with it.
Read next#
Quorum, Synchronous replication, Network partition, Leaderless replication