Geospatial Search is a business critical component for the Uber Marketplace. At its core, it maintains real time state of the drivers and their location and returns the nearby eligible drivers for a given location,

In this talk, We will walk you through how we re-architected Geospatial Search at Uber using Apache Lucene. We will talk about how we handle high volume updates efficiently using a custom built in memory Inverted Index data structure that supports in place term level updates versus the traditional insert after delete paradigm in Lucene. We will share many other optimizations to scale Geospatial search to handle tens of billions of updates per day with subseconds data freshness and serving low latency queries at Uber scale. We talk about how we incorporated ETA based ranking using a machine learning model during the retrieval phase. Lastly we share our experience in running in production to enable mission critical use cases at Uber.