决策地图
可观测性：让运行问题变成可定位的证据
从错误、性能、日志和产品行为中建立可操作的反馈回路。
Recommendation
从错误监控和关键性能路径起步；不要在没有告警责任人的情况下堆叠遥测。
先定义需要回答的生产问题，再选择错误、追踪、日志或分析工具。
范围
用这条边界避免在错误层级解决相邻问题。
包含
- Issue-centric exception and crash grouping
- Centralized log collection, retention, search, and operations
- Application tracing, transaction performance, and service diagnostics
- Host, VM, container, Kubernetes, cloud-resource, process, and system metrics
- External availability and synthetic checks
- Product interaction and identity analysis
- Web acquisition and traffic analysis
- Privacy-controlled individual session replay
不包含
- Product-facing accountable action ledgers, which belong to Security Audit Logs
- Feature rollout control, which belongs to Feature Flags
- Incident status communication
- LLM and agent-specific traces, evaluations, and cost semantics, which belong to AI Observability
- Feature Monitoring as an independent Task
重要工具
这些工具代表此决策领域内不同路线；请先打开相关任务页面，再将任何工具视作默认。
- Sentry （英文页面）
- Better Stack （英文页面）
- Rollbar （英文页面）
- BugSnag （英文页面）
- Datadog （英文页面）
- Axiom （英文页面）
- Grafana Cloud （英文页面）
- New Relic （英文页面）
- Honeycomb （英文页面）
- Prometheus （英文页面）
- UptimeRobot （英文页面）
- Checkly （英文页面）
- PostHog （英文页面）
- Amplitude （英文页面）
- Mixpanel （英文页面）
- Heap （英文页面）
- Google Analytics （英文页面）
- Plausible （英文页面）
- Umami （英文页面）
- Cloudflare Web Analytics （英文页面）
- LogRocket （英文页面）
- Fullstory （英文页面）
- Microsoft Clarity （英文页面）
- OpenReplay （英文页面）
决策顺序
先完成基础归属判断，再增加专门能力。
Is the question about system operation or user behavior?
Separate operational telemetry from behavioral analytics before comparing suites that may expose both.
Error Monitoring Application Logging Application Performance Monitoring Infrastructure Monitoring Product Analytics
Is the operational unit a grouped failure, a searchable event, an instrumented request path, or an external symptom?
Choose the signal that matches triage, investigation, transaction diagnosis, or availability verification.
Error Monitoring Application Logging Application Performance Monitoring Infrastructure Monitoring Uptime & Synthetic Monitoring
Is the behavioral unit an identified product event or an acquisition and traffic session?
Product Analytics centers on product interaction and identity; Web Analytics centers on source, medium, campaign, and site traffic.
Does aggregate data leave an individual interaction unexplained?
Evaluate Session Replay only when qualitative playback is justified and consent, masking, retention, and access constraints are defined.
决策分组
每个任务都对应一个独立的用户决策，而非产品功能标签。
System health and failure signals
These Tasks choose grouped failures, searchable event records, application request traces and performance, or external availability symptoms.
User behavior and experience signals
These Tasks choose product-event analysis, acquisition and site measurement, or qualitative playback of individual sessions.
常见混淆
这些边界避免相邻任务被压缩成一个泛泛的候选清单。
- Error Monitoring and Logging are treated as interchangeable event stores.
- Error Monitoring groups similar exception events into issues; Logging preserves and queries a broader event history.
- Uptime Monitoring and Application Performance Monitoring are treated as the same health signal.
- Uptime Monitoring executes black-box checks against externally visible behavior; APM instruments application internals and request paths.
- Application Performance Monitoring and Infrastructure Monitoring are treated as one generic monitoring decision.
- APM owns application request and service behavior; Infrastructure Monitoring owns host, container, cluster, cloud-resource, process, and system metrics.
- Product Analytics and Web Analytics are treated as one generic analytics decision.
- Product Analytics centers on product events and identity; Web Analytics centers on acquisition, traffic, source, medium, and campaign.
- Feature Monitoring is treated as a retained independent Task.
- Adoption belongs to Product Analytics, rollout control to Feature Flags, and failures or performance to Error Monitoring and Application Performance Monitoring.
暂缓引入
不要在产品真正需要之前引入专门系统。
Aggregate events and operational telemetry already answer the question without individual interaction playback.
Replay adds privacy, masking, retention, storage, and access-control obligations.
The product has no meaningful website acquisition, referrer, or campaign decision.
Product-event analysis can proceed without adding a separate acquisition and traffic measurement surface.
The selected hosting platform already exposes sufficient resource health for the current workload and no independent metric, topology, retention, or alerting requirement exists.
A separate infrastructure metrics platform should follow a concrete resource-health and operating-ownership need.
官方资源
支撑此分类边界的标准与一手文档。
来源
此决策地图使用的主张级参考资料。
- 1Signals
OpenTelemetry · Accessed Official
- 2Grouping Issues
Sentry · Accessed Official
- 3Guide to Computer Security Log Management
NIST · Accessed Official
- 4Trace Context
W3C · Accessed Official
- 5Synthetic Monitoring introduction
Grafana Labs · Accessed Official
- 6Set up events
Google Analytics · Accessed Official
- 7Send User IDs
Google Analytics · Accessed Official
- 8Traffic-source dimensions
Google Analytics · Accessed Official
- 9Frequently asked questions
Microsoft Clarity · Accessed Official
- 10Data retention
Microsoft Clarity · Accessed Official
- 11Prometheus overview
Prometheus · Accessed Official