CSE Roadmap

Core tier · about 2 hours · 5 parts

Part 4 of 5 · 2 min read

Synchronous and asynchronous followers

The same latency-versus-loss-risk trade-off from Level 8 applies per follower, and a leader can mix both kinds at once.

Synchronous and asynchronous followers

Think of the head chef waiting for written confirmation from certain trusted cooks before considering a recipe change finalized. With other cooks, the chef simply calls the change out loud and trusts they will catch up eventually.

In plain words: exactly as covered for a single standby in Level 8, each follower in a single-leader setup can be synchronous, with the leader waiting for that follower's confirmation before considering a write committed, or asynchronous, with the leader not waiting at all. A single leader can even mix both: some followers synchronous for strong durability guarantees, others asynchronous purely for read scaling, where losing a small amount of a lagging follower's most recent data in a rare failure would not be a serious problem.

A worked example. A leader with five followers might configure one nearby follower as synchronous, guaranteeing no data loss on failover involving that follower, while the other four remain asynchronous, purely to spread out read traffic without adding commit latency for every write.

The key lines. No new SQL; which followers are synchronous versus asynchronous is a per-follower replication configuration setting, not a per-statement choice.

Synchronous follower, asynchronous follower

A synchronous follower requires the leader to wait for its confirmation before a write is considered committed; an asynchronous follower receives changes without the leader waiting, exactly the trade-off covered for a single standby in Level 8, now applied per follower.

Example: waiting for written confirmation from certain trusted cooks before finalizing a change, while simply calling a change out to others, mirrors mixing synchronous and asynchronous followers.

Remember this

Single-leader replication is Level 8's single standby generalized to many followers: one leader for all writes, any number of followers for reads, each follower independently synchronous or asynchronous using the exact trade-off covered there. The next page covers what happens to a follower that has fallen behind.

Quick notes

  • Single-leader replication directs every write to one leader, which streams changes to one or more followers.
  • Reads can be served from the leader or any follower, since reading never conflicts with anything.
  • Writes must go through the leader specifically, giving every change one unambiguous point of origin.
  • Each follower can independently be synchronous or asynchronous, using the same trade-off covered in Level 8.
  • A single leader can mix synchronous and asynchronous followers for different purposes at once.