Paxos And Read Consistency

In the last post we did a deep-dive into gaining consensus in the ordering of writes using Paxos. We explored how this could be used to replicate a stream of commands to create a set of datastore backups with automated failover. We can upgrade that model with compare-and-swap semantics, and add weakly consistent reads from cluster replicas. We can also choose to route reads via the leader yet actually run them on replicas whilst retaining strongly consistent semantics. Read the rest of this entry »