Telusko Docs

Replication is the process of copying and maintaining data across multiple databases or nodes. Instead of storing data in a single location, identical copies are maintained on multiple servers.

Replication improves:

High availability
Fault tolerance
Read performance
Disaster recovery
Data durability

By maintaining multiple copies of data, applications can continue operating even if one or more database servers become unavailable.

Benefits_of_Replication

Types of Replication

There are three major replication strategies:

Single Leader Replication
Multi-Leader Replication
Leaderless Replication

Single Leader Replication

In Single Leader Replication, one database acts as the Leader (Primary), while the remaining databases act as Followers (Replicas).

Only the leader accepts write requests.

Followers receive replicated data from the leader.

Architecture

             User
               │
               ▼
            Application
               │
               ▼
         Leader Database
          /     |      \
         ▼      ▼       ▼
   Follower1 Follower2 Follower3

Flow:

User → Server → Leader → Followers

Asynchronous Replication

In Asynchronous Replication, the leader immediately responds to the client after updating its own data.

Followers are updated later in the background.

How It Works

The leader completes the write and responds to the user.
Followers update in the background.
If followers fail, they sync again once recovered.

Advantage: Faster response, no resource blocking.

Disadvantage: Consistency is not guaranteed; data can be temporarily inconsistent or corrupted if followers lag behind.

Synchronous Replication

In Synchronous Replication, the leader waits until all follower nodes acknowledge the write before responding to the client.

How It Works

The leader sends a write request to all followers.
The leader waits until all followers return an OK response.
Only then does it confirm success to the user.

Aspect	Advantage	Disadvantage
Data Consistency	Followers are always up-to-date.	Longer response time.
Failover	Any follower can become the leader (all followers have the latest data).	Blocks resources while waiting for confirmations.

In practice: Asynchronous replication is preferred for most use cases due to faster responses and no resource blocking.

Adding a New Follower Node

Create a snapshot of the data from the leader (or an existing follower).
Load the snapshot into the new node.
Establish a connection between the leader and the new node for ongoing replication.

Leader Failure - Electing a New Leader

When the leader fails, a new leader must be elected:

Identify the most up-to-date follower - check timestamps to determine which follower has the latest data.
Inform all other nodes about the newly elected leader.

Tools like FSImage (full metadata snapshot) and EditLogs (incremental change logs) are used for data replication and recovery.

Multi-Leader Replication

Multiple data centers each have their own leader and follower nodes. Leaders communicate with each other across data centers in addition to their own followers.

Key Points:

Each leader communicates with its followers and with leaders of other data centers.
Improves performance and fault tolerance, if one leader fails, others continue to serve.
Enables parallel writes from multiple leaders.

Architecture

 Data Center A                 Data Center B

     Leader A  ◄────────────►  Leader B
      /   \                     /   \
     ▼     ▼                   ▼     ▼
Follower Follower          Follower Follower

Conflict Problem

When two leaders write to the same data simultaneously, conflicts arise.

Example

Leader A

File Name = A

──────────────

Leader B

File Name = B

Both users receive successful responses.

During synchronization:

A ≠ B

Conflict detected.

Conflict Resolution Strategies

Strategy	Description
Last Write Wins	The value with the most recent timestamp becomes the final value.
Leader Priority	Leaders are assigned priorities, and the higher-priority leader's update is retained.
User Resolution	The application presents the conflict to the user, allowing them to choose the correct version.

Leaderless Replication

Leaderless Replication removes the concept of a primary database.

Every node can receive both read and write requests.

A write succeeds only when enough nodes acknowledge the update.

Architecture

          User
            │
            ▼
      Distributed Cluster

      Node A
      Node B
      Node C
      Node D
      Node E

No single node acts as the leader.

Quorum

A Quorum defines the minimum number of nodes that must successfully participate before an operation is considered successful.

For a cluster containing n nodes:

Quorum > n / 2

Operation	Quorum Requirement
Read	More than n/2 nodes must respond
Write	More than n/2 nodes must confirm

Where:

n = Total Number of Nodes

This ensures that at least one node in both read and write operations has the latest data, maintaining consistency across the system.

Summary

Replication is the process of maintaining multiple copies of data across different nodes or databases to improve availability, fault tolerance, read performance, and disaster recovery.
Single Leader Replication uses one primary node to handle all write operations, while follower nodes replicate the data using either Asynchronous or Synchronous Replication.
Asynchronous Replication prioritizes performance by acknowledging writes immediately, whereas Synchronous Replication prioritizes strong consistency by waiting for follower confirmations.
Multi-Leader Replication enables multiple leaders to accept write requests across different locations, improving availability and scalability while introducing the possibility of write conflicts.
Conflict Resolution in multi-leader systems can be achieved using Last Write Wins, Leader Priority, or User Resolution strategies.
Leaderless Replication eliminates the primary node and uses Quorum-Based Reads and Writes to maintain consistency across distributed nodes.
The choice of replication strategy depends on the application's requirements for consistency, availability, latency, fault tolerance, and scalability.

Written By: Muskan Garg

Replication

On this page