Uber's Real-Time Analytics for Mobile App Crashes using Apache Pinot
Original post Real-Time Analytics for Mobile App Crashes using Apache Pinot
Article Summary
Uber's engineering team processes 1,500 mobile app crashes per second—36TB of data daily—and needed real-time analytics to catch issues before they impact millions of users. Here's how they built a system that detects problems in seconds, not hours.
Kriti Dangi, Anil Purohit, and Parijat Bansal from Uber's Developer Platform team detail their migration from Elasticsearch to Apache Pinot for crash analytics. With 11,000 code changes deployed weekly across Uber's mobile apps, the team needed a way to instantly identify which changes cause crashes and alert the right teams before user impact scales.
Key Takeaways
- Pinot reduced query times by 10x compared to Elasticsearch for time range queries
- System handles 1.5M errors per second with 45 day retention across all platforms
- Smart sampling stores compressed crash data while maintaining full aggregation accuracy
- Hybrid table architecture combines real time Kafka streams with batch processing
- Migration cut infrastructure costs by 60% while improving query performance
Uber's switch to Apache Pinot for crash analytics delivered 10x faster queries, 60% cost reduction, and real-time detection capabilities that help release managers decide within seconds whether to proceed with deployments or rollback.
Continue reading
This page summarizes “Real-Time Analytics for Mobile App Crashes using Apache Pinot” by Uber, published Nov 2, 2023. All credit for the original work goes to the authors.
