Course contentsShow
← System Design
Lesson 1356 of 1,91934. Data Replication StrategiesFree lesson

Asynchronous Replication: Speed and Risk

Why async replication provides better performance but risks data loss if the primary fails before propagation.

Asynchronous Replication: Speed and Risk

What You'll Learn

You'll understand why asynchronous replication is faster than synchronous replication but introduces the risk of data loss during primary failures.

How Asynchronous Replication Works

In asynchronous replication, the primary database acknowledges a write operation immediately after storing it locally—without waiting for replicas to confirm they've received the data. The primary then propagates changes to replicas in the background.

Think of it like sending postcards: you drop them in the mailbox and walk away (write confirmed). You don't wait at the post office to verify each postcard reached its destination. Fast for you, but if the post office burns down before pickup, those postcards are lost.

The Performance Advantage

Because writes don't wait for network round-trips to replicas, asynchronous replication offers:

  • Lower write latency – Users see faster response times
  • Higher throughput – The primary isn't bottlenecked by the slowest replica
  • Better geo-distribution – Distance to replicas doesn't slow down writes

The Risk: Replication Lag

The critical tradeoff is replication lag—the time window between when data is written to the primary and when replicas receive it. If the primary fails during this window:

  • Data loss occurs – Recent writes haven't reached replicas yet
  • Failover is lossy – Promoting a replica means accepting data loss
  • Recovery Point Objective (RPO) is non-zero – You lose seconds/minutes of data

This is acceptable for many applications (social media feeds, analytics) but unacceptable for financial transactions or critical records.

Key Takeaway:
Asynchronous replication maximizes performance and availability by not blocking writes, but accepts the risk that a primary failure before replication completes will result in permanent data loss of recent writes.