Linearizability

Introduction

In a distributed database, if we have multiple replicas that converge over time and if clients request data from multiple replicas, we might get different results.

The concept behind linearizability revolves around providing abstraction of a single copy of the database value to the client. By providing this abstraction, we can ensure that all the operations on it are atomic and clients don't worry about distributed replicas.

In a linearizable system, when one client completes the write operation, all other clients will see the same value irrespective of the replica it queries. Linearizability is also called strong consistency or atomic consistency.

Why linearizability?

To understand why linearizability is required, let’s first understand a system which is nonlinearizable.

In the above illustration

• At time T1, the client issues a write request to the leader node, updating its value to 1.

• At time T2, the client issues a read request to the leader ...