可观测性是通过系统对外输出的指标、日志与链路信号推断其内部状态的能力，核心是在没有预设答案的前提下回答未知问题，能力包括全面埋点、高基数维度支持与多信号关联分析。它解决的是故障根因只能靠现场猜测与人工翻查的问题，并不解决埋点缺失或监控只覆盖已知故障场景的问题。
可观测性的核心是回答未知问题，它可以帮助企业实现以下目标：
让未知故障也能被定位。系统内部状态被充分暴露，出现没见过的现象时也能通过信号组合推断出原因。埋点覆盖入口、处理与出口三个位置，单个请求的完整路径被记录，故障时无需猜测哪一段耗时异常。
让排查从猜测变成检索。结构化的指标与链路让工程师按条件筛选，而不是凭经验逐层翻查日志。指标携带的维度可以任意组合，高基数的标签被限制在可控范围，避免维度爆炸带来的存储成本失控。
让性能问题有量化基线。关键路径的耗时分布被持续记录，劣化在用户投诉之前就能被发现。
让跨系统问题有共同视图。调用链把不同服务串起来，端到端的问题不需要在多个系统间人工对齐时间线。
总之，可观测性可以帮助企业实现故障处理从依赖个人经验到依靠系统信号的转变，缩短平均定位时间，把隐性性能问题提前暴露，降低重复故障的排查成本，让系统在规模扩大时仍然保持可维护。
翻译
Observability is the ability to infer a system’s internal state from the metrics, logs and traces it emits externally. Its core is answering unknown questions without a predetermined answer, with capabilities covering comprehensive instrumentation, high-cardinality dimension support and multi-signal correlation. It solves the problem of root causes being guesswork requiring manual searching, but it does not solve the problems of missing instrumentation or of monitoring that only covers known failure modes.
At the heart of observability is answering unknown questions, which can help companies achieve the following:
Make even unfamiliar failures locatable. Internal state is exposed sufficiently that an unprecedented symptom can still be traced to a cause through combinations of signals.
Turn troubleshooting from guessing into retrieval. Structured metrics and traces let engineers filter by conditions instead of digging through logs from experience.
Give performance issues a quantified baseline. Latency distributions on critical paths are recorded continuously, so degradation is caught before user complaints arrive.
Provide a shared view of cross-system problems. Trace context links different services together, so end-to-end issues no longer require manually aligning timelines across systems.
In short, observability can help enterprises shift fault handling from depending on personal experience to relying on system signals, shorten mean time to resolution, expose hidden performance problems earlier, reduce the cost of investigating recurring faults, and keep systems maintainable as they scale.
