Designing a Distributed Key-Value Store

I’m building a simplified distributed key-value store that mimics how large-scale systems like Amazon DynamoDB or Apache Cassandra distribute and manage data across multiple nodes.

Each node acts as a small independent server with its own storage engine. Keys are assigned to nodes using a hash function, allowing for sharding and distributed storage. Communication between nodes happens over sockets.

graph TD
  Client --> NodeA
  Client --> NodeB
  Client --> NodeC

  NodeA["Node A (Port 5000)"]
  NodeB["Node B (Port 5001)"]
  NodeC["Node C (Port 5002)"]

  NodeA -- "Key: foo → put()" --> NodeB
  NodeB -- "Hash(foo) % 3 = 1" --> NodeB
  NodeC -- "get(foo)" --> NodeB

Why this project?

• To learn how distributed data systems route, replicate, and retrieve data.

• To implement inter-process communication between distributed nodes.

• To get hands-on experience with sharding, replication, and fault handling.

Core Concepts:

• Socket communication between multiple servers

• Hash-based sharding: owner_node = hash(key) % num_nodes

• In-memory key-value storage (dict/map)

• Command-line client tools

Planned Stretch Features:

• N-way replication for fault tolerance

• Heartbeat-based failure detection

• Eventual consistency with quorum-based read/write

This is the kind of backend infrastructure powering modern databases and cloud systems. By building it from scratch, I’m diving deep into distributed design and seeing the challenges firsthand.