Skip to content

Lab: Docker Compose ile Cache Observability Stack

Lokal olarak Prometheus + Grafana + custom exporter stack'ini Docker Compose ile kuruyoruz. Production-ready bir başlangıç noktası.

Şükrü Yusuf KAYA
14 min read
Intermediate

Lab #13: Cache Observability Stack

Bu lab uçtan uca bir monitoring stack kuruyor.
Bileşenler:
  • LLM app + Prometheus exporter
  • Prometheus (metrics scrape)
  • Grafana (dashboard)
  • AlertManager (notification)

Adım 1 — docker-compose.yml#

yaml
version: "3"
services:
  app:
    build: ./app
    ports:
      - "8000:8000"
      - "9100:9100"  # Prometheus exporter
    environment:
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
 
  prometheus:
    image: prom/prometheus:latest
    ports:
      - "9090:9090"
    volumes:
      - ./prometheus.yml:/etc/prometheus/prometheus.yml
      - ./alerts.yml:/etc/prometheus/alerts.yml
 
  grafana:
    image: grafana/grafana:latest
    ports:
      - "3000:3000"
    volumes:
      - ./grafana-provisioning:/etc/grafana/provisioning
      - grafana-data:/var/lib/grafana
 
  alertmanager:
    image: prom/alertmanager:latest
    ports:
      - "9093:9093"
    volumes:
      - ./alertmanager.yml:/etc/alertmanager/alertmanager.yml
 
volumes:
  grafana-data:
docker-compose.yml

Adım 2 — prometheus.yml#

yaml
global:
  scrape_interval: 15s
  evaluation_interval: 15s
 
rule_files:
  - "/etc/prometheus/alerts.yml"
 
alerting:
  alertmanagers:
    - static_configs:
        - targets: ["alertmanager:9093"]
 
scrape_configs:
  - job_name: "llm-app"
    static_configs:
      - targets: ["app:9100"]
prometheus.yml

Adım 3 — alerts.yml#

yaml
groups:
  - name: llm_cache
    interval: 30s
    rules:
      - alert: LowCacheHitRate
        expr: |
          sum(rate(llm_cache_hits_total[10m]))
          / (sum(rate(llm_cache_hits_total[10m])) + sum(rate(llm_cache_writes_total[10m])))
          < 0.85
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: "LLM cache hit rate düştü"
          description: "Son 10dk'da hit rate < 85%"
 
      - alert: HighRequestLatency
        expr: histogram_quantile(0.95, rate(llm_request_latency_seconds_bucket[5m])) > 5
        for: 5m
        labels:
          severity: warning
 
      - alert: CostSpike
        expr: rate(llm_request_cost_usd[5m]) > 0.5  # USD/saniye
        labels:
          severity: critical
alerts.yml

Adım 4 — Grafana Provisioning#

yaml
# grafana-provisioning/datasources/prometheus.yml
apiVersion: 1
datasources:
  - name: Prometheus
    type: prometheus
    url: http://prometheus:9090
    access: proxy
    isDefault: true
 
# grafana-provisioning/dashboards/main.yml
apiVersion: 1
providers:
  - name: "LLM Cache"
    type: file
    options:
      path: /var/lib/grafana/dashboards
Grafana datasource + dashboard provisioning

Adım 5 — Kalkış#

bash
# Stack'i başlat
docker compose up -d
 
# Loglar
docker compose logs -f app
 
# Grafana: http://localhost:3000 (admin/admin)
# Prometheus: http://localhost:9090
# AlertManager: http://localhost:9093
Stack'i çalıştır
Production'a Yol
Bu stack'i CI'da reusable Helm chart'a dönüştürebilirsin. Kubernetes'e deploy edip multi-environment monitoring kur.

✓ Pekiştir#

Bir Sonraki Derste#

A/B testing — caching stratejilerini production'da kıyaslama.

Yorumlar & Soru-Cevap

(0)
Yorum yazmak için giriş yap.
Yorumlar yükleniyor...

Related Content

1. Temeller — Context Penceresi Ekonomisi

Bu Eğitim Hakkında ve Prompt Caching Neden Önemli?

Start Learning
1. Temeller — Context Penceresi Ekonomisi

Token Ekonomisi 101: Input vs Output Cost Asimetrisi

Start Learning
1. Temeller — Context Penceresi Ekonomisi

Context Window Evrimi: 4K'dan 1M'a 5 Yılda Ne Oldu?

Start Learning