Kakao Mobility provides taxi, proxy driver, e-bike, shuttle bus, and navigation services all through a single mobile app. We run a Flink pipeline for the services to deliver seamless customer experiences for distance-based fare estimation, usage-based insurance, and trip summary upon user trip completion. The pipeline performs the following tasks in sequence on streams of GPS points from the app:
(1) group GPS points from each trip into a session
(2) trigger aggregation immediately upon trip completion without waiting for timeout to happen
(3) perform map-matching on each session to get the correct sequence of road segments
(4) enrich each road segment with road-specific attributes from Redis
(5) compute results from the enriched data and write to Cassandra

In this talk, we share our distilled experience from developing and operating the session-based pipeline. We highlight why and how we use Redis and Cassandra in the pipeline.