Design a key-value store in HLD - System Design Guide

Problem Statement

When a system stores data in a single file or database table without indexing, retrieving values by keys becomes slow and inefficient as data grows. Also, without proper distribution and replication, the system can become a bottleneck and a single point of failure, causing downtime and data loss.

Solution

A key-value store organizes data as pairs of unique keys and their associated values, enabling fast lookups by key. It distributes data across multiple nodes using consistent hashing to balance load and replicates data for fault tolerance. The system uses in-memory caching for quick access and persistent storage for durability.

Architecture

Client

→Load Balancer

↓

Node 1

(In-Memory +

↓

Replication

Replication

This diagram shows a client sending requests through a load balancer to a distributed key-value store cluster. The cluster consists of multiple nodes with in-memory caching and persistent storage. Replication and consistency mechanisms ensure data durability and availability.

Trade-offs

✓ Pros

→

Fast data retrieval by key due to direct indexing.

→

Scalable by adding more nodes with consistent hashing.

→

High availability through data replication across nodes.

→

Flexible data model supporting any value type.

✗ Cons

→

Limited query capabilities beyond key-based lookups.

→

Complexity in managing data consistency across replicas.

→

Potential data loss if replication and persistence are misconfigured.

Use when your application requires extremely fast read/write access by key, handles large volumes of data distributed across servers, and needs high availability with fault tolerance.

Avoid if your application requires complex queries, relational data integrity, or transactions involving multiple keys frequently.

Real World Examples

Amazon

Amazon DynamoDB uses a key-value store model to provide low-latency access to user session data and shopping cart information at massive scale.

Netflix

Netflix uses Cassandra, a distributed key-value store, to store user viewing history and preferences with high availability and fault tolerance.

Uber

Uber uses key-value stores like Redis for caching real-time location data to enable fast matching of riders and drivers.

Alternatives

Relational Database

Stores data in tables with fixed schemas and supports complex queries and joins.

Use when: When data relationships and complex queries are critical, and strict ACID transactions are required.

Document Store

Stores data as JSON-like documents allowing flexible schemas and nested data.

Use when: When you need semi-structured data storage with rich querying on document fields.

Wide Column Store

Stores data in tables with flexible columns grouped by row keys, optimized for large-scale distributed storage.

Use when: When you need to store large volumes of sparse data with fast lookups by row key.

Summary

A key-value store organizes data as unique keys mapped to values for fast retrieval.

It scales horizontally by distributing data and replicating it across multiple nodes.

This design suits applications needing quick access by key but not complex queries.