Design visualizations for streaming metrics
Overview
This question evaluates the ability to design observability dashboards and visualize streaming-quality metrics, testing competencies in monitoring, anomaly detection, cohort-based diagnosis, percentile analysis, and system-level troubleshooting within the Analytics & Experimentation category for Data Engineer roles, and it blends conceptual understanding with practical application. It is commonly asked because it reveals a candidate's skill in choosing real-time versus retrospective views, handling seasonality and smoothing, defining diagnostic breakdowns and guardrail metrics, and designing interaction patterns for drill-downs and comparisons.
Design a Monitoring and Diagnosis Visualization for a Video-Streaming Metric
Context
You are building an observability dashboard for a global consumer video product. Choose one critical streaming quality metric and design how you would visualize it to monitor health and diagnose issues.
Assume you have standard playback events (play_attempt, start, first_frame, rebuffer, stop), client metadata (device, OS, app version, network), and infrastructure metadata (CDN/POP, ISP, region). The dashboard should support both real-time monitoring and retrospective diagnosis.
Task
Pick one critical metric (e.g., start-failure rate, rebuffering ratio, time-to-first-frame, hours watched per user) and specify:
-
Primary time-series view
- Time granularity (near real-time vs. daily); zoom behavior
- Smoothing and seasonality handling
- Anomaly detection and bands (baseline and confidence/prediction intervals)
-
Key breakdowns for diagnosis
- Device/OS/app version
- Network type/ISP
- Geography (region/country/city)
- Cohorts (new vs. returning, app release, experiment group)
-
Diagnostic visuals
- Percentile bands where applicable (e.g., p50/p90/p99)
- Geographic heatmap
- Funnel from play attempt → start → watch ≥ X minutes
-
Interaction patterns
- How users drill down, pivot dimensions, and compare baselines
-
Guardrail metrics
- Which additional metrics you would co-display to prevent blind spots
Solution
Login required