Performance Patterns

Caching Strategies

Cache Layers

| Layer | Latency | Eviction | Use Case | |-------|---------|----------|----------| | L1 (in-memory) | <1ms | LRU/TTL | Hot data, session state | | L2 (Redis/Memcached) | 1-5ms | LRU/TTL/TTL+LFU | Shared cache across instances | | L3 (CDN) | 10-50ms | TTL + purge API | Static assets, API responses | | L4 (HTTP cache) | Browser-determined | Cache-Control headers | Browser caching |

Cache Patterns

| Pattern | Description | Use When | |---------|-------------|----------| | Cache-Aside | App checks cache first, then DB, populates cache | General purpose | | Read-Through | Cache library loads from DB on miss | Consistent reads | | Write-Through | Write to cache and DB synchronously | Write consistency critical | | Write-Behind | Write to cache, async write to DB | High write throughput | | Cache Invalidation | Explicit removal on mutation | Stale data unacceptable |

Cache Invalidation Rules

1. Time-to-live (TTL) for all caches.
2. Stale-while-revalidate for tolerance.
3. Versioned cache keys on schema changes.
4. Never cache per-user data without explicit user scope.

Database Optimization

Indexing

| Index Type | Use Case | Trade-off | |------------|----------|-----------| | B-Tree | Equality + range queries | General purpose | | Hash | Exact lookups only | No sorting | | GIN | Array/JSON containment | Write overhead | | GiST | Full-text, geospatial | Build time | | Covering | Avoid table heap access | Storage bloat |

Query Optimization

• Use EXPLAIN ANALYZE / query plan analysis.
• Avoid N+1 queries — use eager loading, batching, or GraphQL dataloaders.
• Paginate all list queries (cursor-based preferred).
• Use connection pooling (max_connections per instance).
• Monitor slow query logs.

Connection Pool Sizing

Pool Size = (max_connections - superuser_reserved) / application_instances
Typical range: 5-30 per instance

Code Profiling

| Phase | Tool (Python) | Tool (Node) | Tool (Go) | |-------|---------------|-------------|-----------| | CPU | cProfile, py-spy | 0x, clinic | pprof | | Memory | memory_profiler, tracemalloc | heapdump | pprof | | I/O | strace, iostat | perf, dtrace | strace | | Async | asyncio debug mode | --trace-event-categories | trace package |

Memory Management

• GC tuning: reduce allocation rate, reuse objects (object pools).
• Detect leaks: heap dumps, growth rate monitoring.
• Avoid: circular references (CPython), closures holding large scopes (JS).
• Use: weak references for caches.

Async Patterns

| Pattern | Description | Example | |---------|-------------|---------| | Async/Await | Coroutine-based concurrency | asyncio, async/await in JS | | Promise/Future | Chainable async operations | Promise.all, Future combinators | | Worker Pool | Fixed-size thread/process pool | concurrent.futures, libuv thread pool | | Event Loop | Single-threaded scheduling | asyncio, Node.js, Tokio | | Actor Model | Isolated stateful processes | Erlang/Elixir, Akka |

Bundle Optimization (Frontend)

| Technique | Savings | Effort | |-----------|---------|--------| | Tree shaking | 30-60% | Low | | Code splitting | 40-80% per route | Medium | | Image optimization | 60-90% | Low (use CDN) | | Font subsetting | 50-80% | Low | | Module federation | Shared deps across apps | High |

Core Web Vitals

| Metric | Target | Impact | |--------|--------|--------| | LCP (Largest Contentful Paint) | <2.5s | Perceived load speed | | FID (First Input Delay) / INP | <100ms / <200ms | Interactivity | | CLS (Cumulative Layout Shift) | <0.1 | Visual stability |

Monitor with Lighthouse CI in CI pipeline, RUM in production.