Real-Time Analytics for Mobile App Crashes using Apache Pinot
Article Summary
Uber deploys 11,000 code changes weekly. How do they catch crashes before users notice? They built Healthline, powered by Apache Pinot.
Uber's engineering team processes 1,500 mobile app crashes per second across iOS and Android platforms. They migrated from Elasticsearch to Apache Pinot to achieve real-time crash analytics at massive scale, handling 36TB of crash data daily with 45-day retention.
Key Takeaways
- Processes 1.5M errors per second with sub-second query response times
- Achieved 4x cost reduction and 10x performance improvement over Elasticsearch
- Hybrid table architecture combines real-time Kafka streams with batch processing
- Custom data flattening and compression handles 10-200KB crash payloads efficiently
- Sampling strategy reduced segment size while maintaining analytical accuracy
Critical Insight
Uber cut crash analytics costs by 75% and improved query speed 10x by migrating to Pinot, enabling release managers to make real-time rollback decisions during canary deployments.
