Metryki i monitoring

Strukturalne logi slog, pełna lista metryk Prometheus i gotowy stack Prometheus + Grafana z dashboardem.

Logi strukturalne#

log/slog w formacie JSON (domyślnie) lub text:

yaml

log:
  level: info
  format: json

Każdy request logowany jest z client IP i request ID:

json

{
  "time": "2026-11-05T12:00:00Z",
  "level": "INFO",
  "msg": "request",
  "request_id": "9f2c1a77b3e4d5a6",
  "method": "GET",
  "path": "/api/users",
  "status": 200,
  "backend": "http://localhost:8081",
  "client_ip": "203.0.113.7",
  "duration_ms": 12
}

Metryki Prometheus#

Metryka	Typ	Etykiety
`proxy_requests_total`	counter	`method`, `route`, `backend`, `status`
`proxy_request_duration_seconds`	histogram	`method`, `route`, `backend`
`proxy_inflight_requests`	gauge	—
`proxy_retries_total`	counter	—
`backend_requests_total`	counter	`backend`
`backend_errors_total`	counter	`backend`
`backend_timeouts_total`	counter	`backend`
`backend_healthy`	gauge	`backend`
`backend_circuit_breaker_open`	gauge	`backend`
`backend_healthchecks_total`	counter	`backend`, `result`
`rate_limit_rejected_total`	counter	—
`rate_limit_errors_total`	counter	—

Ograniczona kardynalność

Etykieta route to pierwszy segment ścieżki, nie pełny URL. Dzięki temu /users/1, /users/2… nie tworzą tysięcy serii czasowych.

Scraping#

bash

curl http://localhost:8080/metrics

yamlprometheus.yml

scrape_configs:
  - job_name: reverse-proxy
    static_configs:
      - targets: ['proxy:8080']

Stack Prometheus + Grafana#

bash

ADMIN_TOKEN=change-me docker compose --profile monitoring up -d

Prometheus: http://localhost:9090
Grafana: http://localhost:3000 (admin / GRAFANA_ADMIN_PASSWORD)

Dashboard Reverse Proxy jest provisionowany automatycznie z deploy/monitoring/grafana/dashboards/ i pokrywa:

request rate i latency percentyle,
error rate i upstream traffic split,
circuit breakery i in-flight requests,
wyniki health checków.

Przykładowe zapytania PromQL#

Request rate per backend#

text

sum by (backend) (rate(proxy_requests_total[5m]))

p95 latencji#

text

histogram_quantile(0.95, sum by (le, route) (rate(proxy_request_duration_seconds_bucket[5m])))

Error rate#

text

sum(rate(backend_errors_total[5m])) / sum(rate(backend_requests_total[5m]))

Otwarte breakery#

text

backend_circuit_breaker_open == 1

Odrzucone przez rate limit#

text

rate(rate_limit_rejected_total[5m])

Alerting — przykłady reguł#

yamlalerts.yml

groups:
  - name: reverse-proxy
    rules:
      - alert: BackendDown
        expr: backend_healthy == 0
        for: 1m
        labels:
          severity: critical
        annotations:
          summary: "Backend {{ $labels.backend }} jest unhealthy"

      - alert: CircuitBreakerOpen
        expr: backend_circuit_breaker_open == 1
        for: 30s
        labels:
          severity: warning

      - alert: HighErrorRate
        expr: |
          sum(rate(backend_errors_total[5m]))
          / sum(rate(backend_requests_total[5m])) > 0.05
        for: 5m
        labels:
          severity: critical

/metrics jest niechronione

Endpoint metryk jest celowo nieuwierzytelniony (prostota scrapowania). Jeśli proxy jest wystawione do internetu, ogranicz /metrics na warstwie sieciowej (firewall, sieć wewnętrzna).