irahardianto Resource and Memory Management
1. Always Clean Up
Resources: files, connections, locks, semaphores, memory, OS handles, GPU.
Clean up in ALL paths: success, error, early return.
Language patterns: Go=defer, Rust=Drop(RAII), Python=with, TS=try/finally, Java=try-with-resources.
2. Timeout All I/O
- Network: 30s default, 5-10s interactive
- DB queries: 10s default, per-query complexity
- Files: timeout on network FS
- MQ: configurable, avoid indefinite blocking
3. Pool Expensive Resources
- DB connections: pool 5-20 per instance
- HTTP: keep-alive reuse
- Threads: CPU count (CPU-bound) or I/O wait (I/O-bound)
Pool config: min 5 (warm), max 20-50, idle timeout 5-10 min, validate before use, monitor utilization/wait/timeout.
4. Prevent Leaks
Leak = acquire + never release → exhaust system resources. Detect: monitor fd/connection/memory over time, long-duration tests, leak tools (valgrind, ASan, heap profilers). Prevent: use language cleanup patterns, never rely on manual alone.
5. Handle Backpressure
Problem: producer > consumer → unbounded queue → OOM.
Solutions: bounded queues, rate limiting, flow control, circuit breakers, drop/reject (fail fast > crash).
Memory by Language
GC'd (Go, Java, Python, JS): auto-freed, still must release non-memory resources, GC pauses in latency-sensitive, profile for retained refs.
Manual (C/C++): malloc/free, RAII in C++, prefer smart pointers.
Ownership (Rust): compiler-enforced safety, no GC, no manual mgmt. Arc/Rc for shared ownership.
Related
- Concurrency Mandate GEMINI.md § Concurrency and Threading Mandate
- Concurrency Principles @.gemini/skills/concurrency-and-threading-principles/SKILL.md
- Error Handling GEMINI.md § Error Handling Principles