Chapter 12

1. In a leader-follower configuration, the nodes hold an election and elect a leader for the cluster. All reads and writes go through the cluster leader, while the other nodes monitor the leader and automatically hold a new election if the leader becomes unavailable. As the name implies, in a multi-master configuration, the cluster has several master nodes and each of the master nodes can serve both read and write requests. The master nodes implement some form of distributed consensus algorithm (Raft, Paxos, and so on) to ensure that they always share the same view of the cluster's state.

2. When implementing the checkpoint strategy, workers are periodically asked by the master to persist their current state to durable storage. If this operation succeeds, a new checkpoint is created. If a worker crashes or becomes unavailable, the master will request for the remaining...