How Spotify Works: System Design Deep Dive

Spotify's Technical Challenge

Spotify serves over 500 million users across 180+ markets, streaming from a catalog of 100+ million tracks. The system must handle real-time streaming, personalization at scale, and seamless playback across devices.

Core Architecture Overview

Spotify operates on a microservices architecture with 1,800+ services. Key components include:

1. Audio Delivery System

• Content stored in cloud storage (Google Cloud)
• Multiple audio formats and quality levels
• Chunked delivery for adaptive streaming
• Edge caching for popular content

2. Backend Services

• Microservices communicate via gRPC and HTTP
• Event-driven architecture using Kafka
• Service mesh for traffic management
• Kubernetes for container orchestration

The Music Streaming Pipeline

When You Press Play:

1. **Client Request**: App requests a track via the API Gateway
2. **Access Control**: Verify user subscription and geographic rights
3. **Audio Resolution**: Determine best quality based on network/subscription
4. **CDN Routing**: Find nearest edge server with the content
5. **Chunked Streaming**: Deliver audio in small chunks for buffering
6. **Playback Analytics**: Track listening for royalties and recommendations

Adaptive Bitrate Streaming:

Spotify uses adaptive streaming similar to HLS/DASH:

• Multiple quality levels: 24kbps to 320kbps
• Client monitors bandwidth continuously
• Seamless quality switching without interruption
• Buffer management for smooth playback

Personalization at Scale

Spotify's recommendation engine is legendary. Key systems:

Discover Weekly Pipeline:

1. Collaborative filtering on listening history
2. Natural language processing on playlist names and blog posts
3. Audio analysis using deep learning
4. Candidate generation → ranking → filtering

Real-time Personalization:

• Listening events stream to Kafka
• Flink processes events in real-time
• User taste profiles updated continuously
• Home page personalized on every visit

Data Infrastructure

The Numbers:

• 100+ petabytes of data
• 5 million events per second
• Thousands of data pipelines
• Real-time and batch processing

Technology Stack:

• Google Cloud Platform for compute/storage
• BigQuery for analytics
• Apache Beam for data processing
• Kafka for event streaming
• Cassandra for user data

Handling Scale

Caching Strategy:

• CDN caching for audio files
• Redis for session data and hot metadata
• Application-level caching for API responses
• Client-side caching for offline playback

Database Choices:

• PostgreSQL for relational data
• Cassandra for high-write workloads
• BigTable for time-series data
• Elasticsearch for search

Key Technical Decisions

1. Microservices Over Monolith

Spotify's scale requires independent scaling and deployment of services.

2. Event-Driven Architecture

Decouples services and enables real-time processing at scale.

3. Cloud-Native Infrastructure

Kubernetes and GCP provide flexibility and scalability.

4. Investment in ML Infrastructure

Custom ML platform enables rapid experimentation and deployment.

System Design Interview Tips

When designing a music streaming service:

Core Features to Address:

• Audio streaming with adaptive quality
• Search and discovery
• Playlist management
• Offline playback
• Social features

Key Trade-offs:

• Latency vs quality (adaptive streaming)
• Consistency vs availability (playlist sync)
• Personalization vs privacy
• Storage vs compute (caching decisions)

Metrics to Consider:

• Time to first byte for playback
• Skip rate (user satisfaction)
• Recommendation click-through rate
• App crash rate

Spotify's architecture shows how microservices, event-driven design, and heavy investment in ML infrastructure can deliver a world-class user experience at massive scale.