Scaling Uber Metric System from Elasticsearch to Pinot

At Uber, we’ve been using realtime system to support time-sensitive critical use cases for years, including Gairos, which was initiated in the Marketplace Org and then widely used across the company since 2014, and uMetric, which has emerged rapidly since 2020.

Continuous effort has been spent toward the reliability and performance of these realtime platforms, to cope with traffic growth, increasing number of users, different varieties of use cases, and following work such as operation cost, resource planning, and optimization feature development. In this presentation we'll share the things we did right to solve these challenges, including fully replace Elasticsearch with Apache Pinot as the realtime storage of our ecosystem.