As stated earlier, all data in Cloud Spanner is replicated across multiple data centers according to the instance configuration. This creates a need for nodes in different data centers to reach a consensus. Spanner solves this using a Paxos consensus algorithm, similar to Megastore and Datastore. The Paxos consensus algorithm allows consensus to be reached in very unreliable environments, and is ideal for maintaining consensus across large bodies of data.
Cloud Spanner organizes nodes into Paxos groups, where a member of the group exists in each zone that hosts a replica of the data. A single instance may have many Paxos groups, and the number of Paxos groups scales with the number of nodes in an instance. Each Paxos group is ...
