Métricas RED e USE: os frameworks que cobrem 90% dos casos

# app.py — instrumentação RED com prometheus_client
from prometheus_client import Counter, Histogram
import time

REQUESTS = Counter(
    "http_requests_total",
    "Total de requisições HTTP",
    ["method", "endpoint", "status"],   # cardinalidade controlada
)

DURATION = Histogram(
    "http_request_duration_seconds",
    "Latência de requisições HTTP",
    ["method", "endpoint"],
    buckets=(0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10),
)

async def middleware(request, call_next):
    start = time.time()
    status = "500"
    try:
        response = await call_next(request)
        status = str(response.status_code)
        return response
    finally:
        elapsed = time.time() - start
        REQUESTS.labels(request.method, request.url.path, status).inc()
        DURATION.labels(request.method, request.url.path).observe(elapsed)

# Queries PromQL que derivam as 3 métricas RED

# Rate (req/s nos últimos 5min)
sum by (endpoint) (rate(http_requests_total[5m]))

# Errors (% erros 5xx)
sum by (endpoint) (rate(http_requests_total{status=~"5.."}[5m]))
  / sum by (endpoint) (rate(http_requests_total[5m]))

# Duration p99
histogram_quantile(0.99,
  sum by (endpoint, le) (rate(http_request_duration_seconds_bucket[5m]))
)

# dashboard-checkout-api.yaml — estrutura de um dashboard production-grade

panels:
  - title: "Traffic (RED:Rate)"
    query: |
      sum by (endpoint) (rate(http_requests_total{service="checkout"}[5m]))

  - title: "Error Rate (RED:Errors, Golden:Errors)"
    query: |
      sum(rate(http_requests_total{service="checkout",status=~"5.."}[5m]))
      / sum(rate(http_requests_total{service="checkout"}[5m]))
    alert: "> 1% por 5min"

  - title: "Latency p50/p95/p99 (RED:Duration, Golden:Latency)"
    queries:
      - histogram_quantile(0.50, sum by (le) (rate(http_request_duration_seconds_bucket[5m])))
      - histogram_quantile(0.95, sum by (le) (rate(http_request_duration_seconds_bucket[5m])))
      - histogram_quantile(0.99, sum by (le) (rate(http_request_duration_seconds_bucket[5m])))

  - title: "CPU Saturation (USE)"
    query: |
      sum by (pod) (rate(container_cpu_usage_seconds_total{pod=~"checkout-.*"}[5m]))
      / sum by (pod) (kube_pod_container_resource_limits{resource="cpu",pod=~"checkout-.*"})
    alert: "> 85% por 10min"

  - title: "DB Connection Pool Saturation (USE)"
    query: |
      pool_connections_active{pool="checkout_db"}
      / pool_connections_max{pool="checkout_db"}
    alert: "> 90%"

  - title: "DB Query p99 (RED aplicado ao DB)"
    query: |
      histogram_quantile(0.99,
        sum by (query_type, le) (rate(db_query_duration_seconds_bucket[5m]))
      )
    alert: "> 500ms por 10min"

  - title: "Queue Saturation (SQS consumer lag)"
    query: aws_sqs_approximate_number_of_messages_visible{queue="checkout-events"}
    alert: "> 10000 por 15min"