51d: Eventual Consistency

In the leader-follower data systems, if we don’t pin clients to the followers, the clients are able to read data through different followers. The followers might be out-of-date. One follower might be behind the other due to asynchronous data replication. So when a client requests data first from one and then another, the result is different. But as soon as the write operation stops, all followers will eventually be consistent. If the system doesn’t require high consistency, this model is good. However, building systems on top of eventually consistent systems is difficult.

• Related Note(s):
  • DDIA: What is Replication in Distributed Systems;
  • Understanding How Database Transactions Work;
• Source(s):
  • Understanding Distributed Systems;
  • Software World #24: Understanding Distributed Systems with Roberto Vitillo;