Prometheus 是以时间序列方式采集、存储并对系统指标进行查询与告警的开源监控组件，核心是拉取模型下的指标存储与 PromQL 查询能力，能力包括多维标签建模、规则告警与长期存储对接。它解决的是业务指标没有统一采集口径、监控与业务系统强耦合的问题，并不解决指标本身定义错误或数据采集频率过低的问题。
Prometheus 的核心是指标建模与查询，它可以帮助企业实现以下目标：
让指标带上可切分的维度。服务名、实例、环境与版本作为标签，同一指标可以按任意维度聚合与对比。同一指标在不同标签组合下的数值可以相互印证，异常往往先在某个维度的偏离上显现。
让告警基于查询结果而非静态阈值。记录规则与告警规则用 PromQL 表达，复杂条件可以组合表达而不需要写代码。规则评估在服务端周期执行，触发后直接对接通知渠道，告警链路不依赖人工查看面板。
让监控与业务解耦。Prometheus 主动拉取目标，服务端不需要嵌入 SDK，减少对业务进程的影响。
让历史数据可用于回溯分析。指标按时间存储，容量变化与故障前后的曲线可以长期对比分析。
总之，Prometheus 可以帮助企业实现可观测数据从各系统自建采集到统一标准模型的转变，提升指标口径一致性与故障发现速度，降低监控体系对业务代码的侵入，让性能数据从临时排查工具变成可长期积累的分析资产。
翻译
Prometheus is an open-source monitoring component that collects, stores, queries and alerts on system metrics in a time-series manner. Its core is metric storage under a pull model together with PromQL query capability, with capabilities covering multi-dimensional label modeling, alerting rules and long-term storage integration. It solves the problems of business metrics lacking a unified collection standard and of monitoring being tightly coupled to business systems, but it does not solve the problems of incorrectly defined metrics or insufficient collection frequency.
At the heart of Prometheus is metric modeling and querying, which can help companies achieve the following:
Make metrics carry sliceable dimensions. Service name, instance, environment and version act as labels, so the same metric can be aggregated and compared along any dimension.
Base alerts on query results rather than static thresholds. Recording and alerting rules are expressed in PromQL, letting complex conditions combine without writing code.
Decouple monitoring from business systems. Prometheus pulls targets actively, so servers need no embedded SDK and business processes are less affected.
Make historical data available for retrospective analysis. Metrics are stored over time, so capacity changes and pre-incident curves can be compared across long periods.
In short, Prometheus can help enterprises shift observability data from each system building its own collection to one unified standard model, improve metric consistency and fault detection speed, reduce how deeply monitoring intrudes on business code, and turn performance data from an ad-hoc troubleshooting tool into a long-term analytical asset.
