İçeriğe geç
Tüm roadmap'e dön
topiccore

Rate Limiting & Queue Management

Vendor rate limit + concurrent agent limit + queue with backpressure.

2 saat

Production agent sistemde 3 katman:

  1. Vendor rate limit (OpenAI/Anthropic) — RPM + TPM quota. Aşılırsa 429.
  2. Concurrent agent limit — eş zamanlı kaç agent çalışabilir (CPU/RAM/cost).
  3. Per-user rate limit — abuse prevention.

Pattern:

  • Token-bucket algoritması (her user için)
  • Redis-based distributed rate limiter
  • Queue (BullMQ / Sidekiq / Temporal) backpressure ile
  • 429 retry with exponential backoff
  • Priority queue (paid user > free, real-time > async)