常见的监控工具有哪些，你如何在它们之间进行选择？

Question

Accepted Answer

监控工具的格局按支柱分类——metrics、logs、traces——加上综合性托管平台。选择归结为**自托管 vs 托管**，由团队规模、预算和规模驱动。

## 按支柱分类的工具

```text
METRICS    Prometheus  → pull-based scraping, time-series DB, PromQL query language
           Grafana     → dashboards on top of Prometheus (and many other sources)

LOGS       ELK         → Elasticsearch + Logstash + Kibana (powerful, heavy to run)
           Loki        → "Prometheus for logs": cheap, indexes labels not full text

TRACES     Jaeger      → distributed tracing, OpenTelemetry-compatible
           Tempo       → trace backend that pairs with Grafana/Loki

ALL-IN-ONE Datadog     → managed metrics + logs + traces + APM in one product
```

## 为什么 Prometheus + Grafana 是常见的开源核心

Prometheus 从它scrapes的目标**拉取** metrics（服务暴露 `/metrics`），将其存储在时间序列数据库中，并使用 **PromQL** 进行查询。拉取方式使目标易于发现和健康检查。Grafana 位于顶层用于可视化。Loki 和 Tempo 将相同的堆栈扩展到日志和追踪，所有内容都可在一个 Grafana UI 中查询。

```promql
# PromQL: per-route error ratio over 5 minutes
sum by (route) (rate(http_requests_total{status=~"5.."}[5m]))
  / sum by (route) (rate(http_requests_total[5m]))
```

## 自托管 vs 托管的权衡

```text
SELF-HOSTED (Prometheus/Grafana/Loki/Tempo)
  ✓ no per-host bill, full control, no data leaves your infra
  ✗ you run, scale, and patch it; storage & HA are your problem

MANAGED (Datadog, Grafana Cloud)
  ✓ no ops burden, integrations out of the box, scales for you
  ✗ costs grow fast with hosts/volume; some vendor lock-in
```

## 如何选择

```text
Small team / tight budget   → self-hosted Prometheus + Grafana (+ Loki)
Fast-moving, ops-light team  → managed (Datadog / Grafana Cloud)
Large scale / strict cost    → self-host the core, but budget for the ops team
```

## 为什么这很重要

没有通用的最佳工具——只有适合你的约束条件的合适工具。一个5人的初创公司尝试运维高可用 Elasticsearch 集群会陷入困境；一家大型企业可能每年在托管平台上花费六位数，而这个平台本可以自托管。了解每个工具的功能以及自托管 vs 托管的权衡，能让你根据团队规模、预算和规模来匹配可观测性，而不是盲目跟风时下流行的工具。