Dashboards

Dashboards Answer Questions

 every panel earns its place by answering a QUESTION:

 "is the service healthy right now?"     → triage view
 "which dependency is misbehaving?"      → drill-down view
 "how are we trending this month?"       → SLO/capacity view

 dashboards built by ACCUMULATION ("add a graph for each
 metric we emit") become wallpaper nobody can read at 3am.
 design top-down from the question, not bottom-up from metrics.

 the three-tier layout that works:
 [row 1] SERVICE HEALTH: golden signals + SLO burn — answers
         "do I need to care?" in five seconds
 [row 2] DEPENDENCIES: downstream health, queue ages, pools
 [row 3] DETAIL: per-endpoint/per-instance for drilling

Design Rules That Survive 3am

 □ FIVE-SECOND RULE: row one answers healthy/not without
   scrolling or thinking; color + threshold lines do the work
 □ ANNOTATE DEPLOYS on time axes (the first question in any
   incident: what changed?)
 □ CONSISTENT LAYOUT across services: same panels, same order,
   same names — muscle memory is a debugging tool; template it
 □ LINKS TO NEXT STEPS: panel → related trace search, runbook,
   dependent-service dashboards (navigation IS diagnosis)
 □ AGGREGATE-FIRST: totals and percentiles up top;
   per-instance heatmaps below for outlier hunts
 □ TIME-RANGE sanity defaults: incident view = last 3h with
   zoom; review view = week; never make people reconfigure mid-fire

Dashboard Hygiene

Rot	Cure
40-panel walls of graphs	question-driven redesign, delete freely
queries nobody remembers	ownership labels + quarterly pruning
broken/empty panels	alert on dashboard-query errors!
copy-paste divergence across teams	templated library, shared variables
vanity metrics front-page	move to appendix or delete

 dashboards rot like code because they ARE code:
 versioned, reviewed, tested (does the query still return?),
 and refactored when questions change. treat a dashboard PR
 like any other engineering artifact.

Interview Framing

“Build the operational dashboard for our payment service” scored shape: three-tier question-driven layout drawn, five-second-rule and deploy annotations emphasized, consistency/template point made (fleet-wide muscle memory), hygiene table condensed, links-to-next-steps navigation included. Dashboard questions test information DESIGN under stress — you’re graded on the 3am user’s experience, not pixel count.