Case Study: Pinterest achieves real-time user engagement insights with MemSQL

How Pinterest Measures Real-Time User Engagement with Spark

Pinterest needed a way to measure and react to real-time user engagement—such as pins, repins, comments, and logins—across its global user base without the latency of MapReduce. The challenge was reliably ingesting high-volume event streams, enriching them with pin and geolocation data, ensuring deduplication, and making the results immediately queryable by analysts and partners.

Pinterest built a pipeline using Kafka → Spark Streaming → MemSQL (via the MemSQL Spark Connector) to filter, enrich, deduplicate, and store events in separate tables while using MemSQL as a key/value cache for reuse. The setup powers real-time visualizations of repins by location, gives analysts SQL access to live engagement metrics, serves as a source of record on AWS, and provides a repeatable, production-ready streaming solution that lets Pinterest detect and respond to trends as they happen.

Pinterest