performance-patterns

Performance Anti-Patterns Reference

N+1 Query Problem

The N+1 problem occurs when code executes N additional queries to fetch related data for N items from an initial query.

Identification:

• Queries inside loops
• Lazy loading of associations during iteration
• GraphQL resolvers fetching per-item

Fix Strategies:

1. Eager Loading: Load related data in initial query
2. Batching: Collect IDs, fetch all at once
3. DataLoader: For GraphQL, batch and cache per-request
4. Denormalization: Store computed/related data together

Severity: HIGH - Scales linearly with data size, causes exponential slowdown

Over-Fetching

Retrieving more data than needed from API or database.

Identification:

• SELECT * queries
• API endpoints returning full objects
• No field selection support
• Loading nested relations by default

Fix Strategies:

1. Field Selection: Only query needed columns
2. Sparse Fieldsets: Support ?fields=id,name parameter
3. GraphQL: Let clients specify exact fields
4. DTOs: Map to response-specific objects

Severity: MEDIUM - Increases bandwidth, memory, serialization time

Under-Fetching

Requiring multiple requests to get needed data.

Identification:

• Waterfall requests (request depends on previous)
• Multiple endpoints for related data
• No include/expand support

Fix Strategies:

1. Compound Endpoints: /users?include=orders
2. GraphQL: Single query for nested data
3. BFF Pattern: Backend aggregates for frontend
4. Parallel Requests: When dependencies allow

Severity: MEDIUM - Increases latency, connection overhead

Missing Pagination

Returning unbounded result sets.

Identification:

• List endpoints without limit
• findAll() without pagination
• No cursor for large datasets

Fix Strategies:

1. Offset Pagination: ?page=1&limit=20
2. Cursor Pagination: ?cursor=abc&limit=20 (better for large sets)
3. Default Limits: Always apply max limit server-side
4. Streaming: For very large exports