Course contentsShow
← System Design
Lesson 237 of 1,9197. Database Scaling: ShardingFree lesson

Sharding Architecture Patterns

Comparing application-level sharding, middleware proxies, and database-native sharding implementations.

Sharding Architecture Patterns

What you'll learn: The three main architectural approaches to implementing database sharding and their trade-offs.

Now that you understand shard mapping and routing, it's time to see where that routing logic actually lives. There are three common patterns for implementing sharding in your system architecture.


Application-Level Sharding

Your application code directly handles all shard routing. The app knows about all shards, performs shard key hashing, and connects to the correct database.

Analogy: Like being your own travel agent—you personally decide which hotel (shard) to book based on the destination (shard key).

Pros: Maximum control, no middleware overhead, easy to customize
Cons: Routing logic scattered across services, harder to change shard strategies later


Middleware Proxy Sharding

A dedicated proxy layer sits between your application and databases. Applications connect to the proxy as if it's a single database; the proxy handles all routing transparently.

Analogy: Like using a travel booking service—you tell them where you want to go, they handle the hotel selection behind the scenes.

Pros: Centralized routing logic, easier schema changes, application stays simple
Cons: Extra network hop, proxy becomes a critical component, potential bottleneck

Examples: ProxySQL, Vitess, Citus


Database-Native Sharding

The database system itself manages sharding internally. You interact with it like a single database, and it distributes data automatically.

Analogy: Like a hotel chain that automatically books you into the nearest property—you don't see the complexity.

Pros: Seamless integration, battle-tested logic, minimal application changes
Cons: Vendor lock-in, less control, sometimes limited customization

Examples: MongoDB sharding, Elasticsearch sharding, Azure Cosmos DB


Key Takeaway: Application-level sharding gives maximum control but scatters logic; middleware centralizes routing but adds a hop; database-native sharding hides complexity but reduces flexibility. Choose based on your team's expertise and control requirements.