Course contentsShow
← System Design
Lesson 1392 of 1,91934. Data Replication StrategiesFree lesson

Monotonic Write Consistency

Guarantee that writes from a single client are applied in the order they were issued.

Monotonic Write Consistency

What You'll Learn

You'll learn how to guarantee that writes from a single client are applied in the order they were issued across replicas.

The Core Problem

Imagine you update your profile photo, then immediately update your bio. Without monotonic write consistency, one replica might see only the bio update while another sees both, or worse—see them arrive out of order. This creates confusion and potential data corruption.

Monotonic write consistency ensures that if a client performs write W1 followed by write W2, any replica that applies W2 has already applied W1. The writes maintain their ordering as they propagate through the system.

How It Differs from Related Guarantees

You've learned about read-your-writes (seeing your own writes) and monotonic reads (not going backward in time when reading). Monotonic write consistency focuses on write ordering from a single client's perspective.

Implementation Approaches

Session Tokens

The client receives a token after each write containing a version/timestamp. The next write includes this token, forcing the system to wait until W1 has propagated before accepting W2.

Sticky Routing

Route all writes from a client session to the same leader/coordinator node, which naturally preserves order.

Sequence Numbers

Assign monotonically increasing sequence numbers to writes from each client. Replicas buffer out-of-order writes until gaps are filled.

Why It Matters

Without this guarantee, operations like:

  • Creating a record then updating it
  • Deleting a file then recreating it
  • Sequential account transactions

...can arrive scrambled, causing application logic to break or data to become inconsistent.

Key Takeaway: Monotonic write consistency prevents your own writes from appearing out of order across replicas, ensuring logical causality is preserved even in distributed replication scenarios.