Observability на собеседовании системного аналитика

Что такое observability

Ability понять internal state системы из external outputs.

Vs monitoring. Monitoring — known unknowns (predefined dashboards). Observability — unknown unknowns (debug new issues).

3 pillars

Logs. Discrete events.

Metrics. Numerical measurements over time.

Traces. Request path through distributed system.

Каждый pillar для different debugging needs.

Logs

Detailed records — что произошло, когда.

{"ts":"2026-05-07T12:00:00Z","level":"INFO","msg":"order_created","user_id":42}

Strengths. Detailed, narrative.

Weaknesses. Volume, slow query без index.

Metrics

Aggregated numerics over time.

http_requests_total{status="200"} 12345
http_request_duration_seconds_p99 0.250

Strengths. Cheap, aggregatable, fast queries.

Weaknesses. No detail на specific request.

Prometheus standard.

Traces

Request through services.

[Client] → [API] → [Auth] → [DB] → [Auth response] → [API] → [Client]
   (50ms total — 30ms в DB)

Strengths. Service-to-service performance.

Weaknesses. Sampling cost; complex setup.

OpenTelemetry — standard.

Tools

Logs: ELK (Elasticsearch + Logstash + Kibana), Loki + Grafana, Splunk, Datadog Logs.

Metrics: Prometheus + Grafana, Datadog, VictoriaMetrics.

Traces: Jaeger, Tempo, Honeycomb.

Combined. Datadog, New Relic, Grafana stack — unified.

В РФ: Open source stack (Prometheus + Loki + Tempo + Grafana) — стандарт.

Связанные темы

• Logging strategy для SA
• Distributed tracing для SA
• SLA SLO SLI для SA
• Health checks для SA
• Подготовка к собесу системного аналитика

FAQ

Это официальная информация?

Нет. Статья основана на индустриальных observability practices.

Тренируйте системный анализ — откройте тренажёр с 1500+ вопросами для собесов.