대시보드 입문: 지표 정의에서 첫 차트까지

Before building a dashboard, write each metric as one sentence: numerator, denominator, time window, data source. This list follows storage, query, visualization, documentation — aiming for one trustworthy chart rather than a screen of attractive ones.

Shortlist（6）

• ClickHouse · 深度导读卡 ¥3.9Apache-2.0 · C++ · ★50.7k

  Columnar storage suits wide aggregation scans; land data where it can be scanned quickly first.

• grafana · 深度导读卡 ¥3.9AGPL-3.0 · TypeScript · ★77.7k

  Visualization and alerting in one place, useful when ops and business metrics share a screen.

• marimo · 深度导读卡 ¥3.9Apache-2.0 · Python · ★23.4k

  Reproducible notebook-style analysis for validating definitions before freezing a dashboard.

• mermaid · 深度导读卡 ¥3.9MIT · TypeScript · ★91.0k

  Describe flows and dependencies as text, making 'where does this number come from' explicit.

• analytics · 深度导读卡 ¥2.9MIT · JavaScript · ★2.7k

  Lightweight traffic analytics feeding first-party traffic definitions into content dashboards.

• netdata · 深度导读卡 ¥4.9GPL-3.0 · Go · ★81.3k

  Keep resource monitoring separate from business metrics so a busy server is not mistaken for a healthy business.

Define, then query

Three teams may compute 'active user' three ways. Write the definition into a versioned document so meetings do not become debates.

One screen, one question

Putting 'is today abnormal' and 'what is the quarterly trend' on one screen obscures both. Split dashboards by usage scenario.

Keep raw data

Aggregates speed up queries, but definition changes require recomputing from raw data. Dashboards that only keep aggregates must rebuild history on every change.

Common pitfalls

Common traps: treating page views as users, concluding from tiny samples, mixed time zones shifting day boundaries, and missing freshness labels.

Alternatives

With small data, a scheduled export plus version history may suffice; a full dashboard platform adds maintenance surface.