Spotify Architecture
Instant playback and personalised discovery for a global catalogue of audio.
On the roadmap:Software Architect · step 5: Learn from the giants
Overview
Spotify separates the data path (audio delivery through caches and CDNs) from the control path (catalogue, accounts, playlists), then feeds listening events into a streaming pipeline that powers recommendations and reporting. This summary is written from public engineering talks and blog posts; see the linked resources for the originals.
Architecture highlights
- Audio is chunked and delivered from CDNs, with a peer-assisted fallback in early versions
- Small autonomous squads own services; the backend grew into many microservices
- Event streams feed analytics, recommendations and royalty reporting
- Recommendations blend collaborative filtering with audio and text signals
Architecture diagram
Key components
Streaming and CDN layer
Delivers audio chunks close to the listener so playback starts quickly
Metadata and catalogue services
Serve track, album, artist and playlist data to every client
Event pipeline
Collects listening events for analytics, billing and model training
Recommendation platform
Builds playlists and home-screen suggestions from listening behaviour
Challenges and trade-offs
Start-up latency
Playback must begin in a fraction of a second, so the first audio chunks are prefetched and cached near users.
Personalisation at scale
Recommendations are recomputed for a huge user base, which needs batch pipelines plus fast online serving.
Team autonomy versus consistency
Many independent services need shared standards for deployment, monitoring and data contracts.
Study: Spotify
Learn how Spotify is built from the people who explain it. Watch first for the picture, then read for the detail.
34 minSpotify ML Question - Design a Recommendation System (Full mock interview)
Aced (formerly Exponent) · 89K views