Design a scalable video platform database
Company: Google
Role: Data Scientist
Category: Data Manipulation (SQL/Python)
Difficulty: Medium
Interview Round: Technical Screen
Design the relational database for a YouTube-like video company. Deliverables: 1) list the core tables with key columns, types, and constraints (users, channels, videos, video_transcodes/qualities, captions, tags, video_tags, views, likes, comments, subscriptions, playlists, playlist_videos, ad_impressions, daily_video_metrics); 2) define primary/foreign keys, uniqueness, and soft-delete and GDPR-compliant deletion strategies; 3) model many-to-many relationships (e.g., videos↔tags, playlists↔videos) and idempotent ingest (avoid duplicate views/likes); 4) include indexing/partitioning (e.g., views partitioned by event_date, video_id; clustered indexes for hot queries), and how you’d support both OLTP and analytics (star schema or read-optimized warehouse tables) without blocking writes; 5) show sample CREATE TABLE DDL for 3–4 critical tables (videos, views, comments, ad_impressions) and explain how you’d query: a) watch-time per video per day, b) top N videos by unique viewers in the last 7 days, c) comments pagination with anti-abuse flags; 6) describe how you’d store multiple renditions (1080p, 4K, HDR) and A/B test assignments for thumbnails.
Quick Answer: This question evaluates relational database design and data engineering competencies—including schema modeling, many-to-many relationships, idempotent ingest, indexing and partitioning, OLTP versus analytics integration, GDPR-compliant deletion strategies, and query formulation—within the Data Manipulation (SQL/Python) domain for a Data Scientist role. It is commonly asked to assess both conceptual understanding and practical application of scalable data architectures, performance tuning, and compliance trade-offs, focusing on the ability to reason about schema choices, read/write optimization, and analytics integration without implementation details.