minions-ai-agents/antigravity_brain_export/knowledge/performance_standards.md

🏎️ Performance Standards (The "Turbo" Protocol)

Audience: Linus Torvalds (The Kernel) & The Architect. Objective: Speed is a Feature. Latency is a Bug.

[!CRITICAL] The Linus Mandate: "I don't care if it's 'clean code'. If it takes 2 seconds to load a dropdown, it's garbage. Optimize it or delete it."

1. ⏱️ Latency Budgets (Hard Limits)

Every endpoint/function falls into a Tier:

• Tier 1 (Interactive): < 100ms (UI interactions, Auto-complete).
• Tier 2 (Standard): < 500ms (Form submissions, Page loads).
• Tier 3 (Batch): < 5s (Reports, Complex calculations).
  • Rule: If it takes > 5s, it MUST be Async (Background Job).

2. 🗄️ Database & IO (The N+1 Killer)

• No Loop Queries: NEVER execute a SQL query inside a for loop.
  • ❌ BAD: for user in users: db.get_profile(user.id)
  • ✅ GOOD: db.get_profiles(user_ids) (Batch fetching).
• *Select : Explicitly select columns. Do not drag 5MB of JSONB if you only need the id.

3. 💾 Memory & Resources

• Generators > Lists: Use yield for processing large datasets. Do not load 1GB CSVs into RAM.
• Container Limits: Respect the 512MB limit set in business_standards.md.
  • Leak Detection: If memory grows linearly over 1 hour, the container must be killed and investigated.

4. ⚡ Caching Strategy

• Read-Heavy Data: Must be cached (Redis/Memcached) with a TTL.
• Cache Stampede: Implement "Stale-While-Revalidate" patterns. Do not let 1000 users hit the DB simultaneously when cache expires.

5. 🐧 The Kernel's Optimization Checklist

Before merging, run the profiler:

• Complexity: Is this algorithm O(n) or O(n^2)? (Nested loops over large data = Reject).
• IO: Count the DB calls. Is it 1 or 100?
• Payload: Check the JSON response size. Is it > 100KB? (Gzip it or paginate it).
• Async: Are we blocking the Event Loop (Node/Python Asyncio) with CPU work?