Katkat
Back

Design a Netflix-like streaming platform

HubSpotSoftware Engineer · Onsite
HardSoftware EngineerOnsiteSystem Design

Last updated

A HubSpot software-engineer onsite system-design question: design a Netflix-like video streaming platform covering catalog browse by genre and curated rails (Recently Added, Top Ten), search, adaptive-bitrate playback across web, mobile and TV, cross-device resume, and an upload-to-streamable ingest pipeline. It evaluates service decomposition, data models and API contracts, transcoding and CMAF/multi-DRM packaging, CDN and caching strategy, session concurrency and consistency, multi-region failover, observability and A/B testing, graceful degradation, and cost control at millions of concurrent viewers. The worked solution sizes egress and heartbeat load, corrects the common concurrency, resume, and cache-TTL mistakes, and marks the MVP cut line.

Question

Design a Netflix-like video streaming platform. Viewers must be able to browse and search a catalog, stream titles with adaptive bitrate (ABR) across web, iOS, Android, and smart TV, and resume playback where they left off across sessions and devices. Content providers (studios, plus user-generated uploads) must be able to upload a video that becomes streamable after processing.

Assumptions you may refine: ~50M DAU with a peak of ~5M concurrent streams; global audience with multi-region deployment; source masters of 1-10 GB producing multiple encoded renditions. Non-functional goals: low startup latency, smooth playback, global availability, fault tolerance, observability, and content protection (at minimum tokenized URLs, plus DRM for licensed content). Be explicit about what you would cut for a first MVP (~100k DAU, three use cases: browse, play, resume) and what is non-negotiable at any scale.

Work through the full design and be ready to cover:

  1. Requirements, scope, and scale assumptions - functional vs non-functional requirements, what is explicitly out of scope, and back-of-the-envelope sizing for egress bandwidth, control-plane QPS, and storage.
  2. Core use cases - browse the catalog by genre and by curated rails such as "Recently Added" and "Top Ten"; play a video at different resolutions on every device type; resume the last play position across sessions and devices.
  3. Service decomposition and high-level architecture - ingestion/transcoding, metadata and catalog, search/indexing, recommendations, playback and ABR, auth/session, DRM licensing, CDN, and how the pieces fit together.
  4. Data models - titles/series/episodes, video assets and the ABR ladder, curated and ranking lists, users/profiles/devices, playback sessions, and watch progress.
  5. API contracts - client-facing (catalog browse, search, playback start, heartbeat, stop, resume/continue-watching) and uploader-facing (multipart upload, complete, status).
  6. Video ingestion and transcoding pipeline - master upload and validation/QC, the ABR ladder decision, HLS/DASH packaging over CMAF, thumbnails and subtitles, DRM packaging, and how a title flips to "ready".
  7. Search, indexing, and recommendations - full-text search over title/tags/cast with faceting and typo tolerance, index freshness, personalized rows, and cold start for a brand-new profile.
  8. Adaptive bitrate streaming and DRM - ABR ladder and segment sizing, HLS/DASH packaging over a shared CMAF segment set, multi-DRM (Widevine/FairPlay/PlayReady) with CENC, signed/tokenized URLs, and per-device player support and fallback.
  9. Auth, profiles, and entitlements - sign-in and sessions, TV device pairing, multiple profiles per account, maturity ratings and parental controls, regional licensing windows, and per-account concurrent-stream limits.
  10. Playback session management and resume - session lifecycle, heartbeats and debounced progress writes, atomic concurrency enforcement, and how resume is restored and conflicts resolved across devices.
  11. "Recently Added" and "Top Ten" - how to define, compute, materialize, and serve these rails, with a small worked ranking example.
  12. Storage, CDN, and caching - object storage vs SQL vs NoSQL choices, geo-replication, multi-CDN and origin shield, manifest vs segment TTLs, and cold starts / cache warming for fast time-to-first-frame.
  13. Scalability, availability, and consistency - partitioning, autoscaling, SLOs and capacity targets at peak concurrency, multi-region topology and failover, and the consistency strategy for catalog, entitlements, rankings, and resume state.
  14. Observability, experimentation, and failure handling - QoE and backend metrics, tracing, A/B testing with guardrails, and graceful degradation (CDN fallback, origin/license failover, partial catalog outage, transcode backlog, write failures).
  15. Security, abuse prevention, and cost optimization - rate limiting, signed-URL rotation, anti-piracy and watermarking, credential-sharing detection, and the levers that control egress and encoding spend at millions of concurrent viewers.