← Lessons

quiz vs the machine

Silver1120

System Design

Semi Synchronous Replication

A practical middle ground that keeps one follower in sync while the rest catch up later.

4 min read · intro · beat Silver to climb

The middle path

Fully synchronous replication is fragile because any slow follower blocks every write. Fully asynchronous risks losing confirmed data. Semi synchronous replication blends the two.

In this scheme, the leader waits for exactly one follower to confirm a write, then acknowledges the client. The remaining followers receive the change asynchronously.

Why this helps

  • It guarantees that every acknowledged write lives on at least two nodes, so a single leader crash loses nothing.
  • It avoids waiting for the slowest follower, since only one confirmation is required.
  • If the synchronous follower falls behind, the system can promote another follower into the synchronous slot to keep going.

The catch

If the one synchronous follower becomes unreachable and no replacement exists, the system must either block writes to preserve the guarantee or quietly fall back to asynchronous, which reopens the data loss window.

Key idea

Semi synchronous replication keeps just one follower in lockstep to gain durability without paying for every node to confirm.

Check yourself

Answer to earn rating on the learn ladder.

1. How many followers must confirm before a semi synchronous write is acknowledged?

2. What durability guarantee does semi synchronous replication provide?