Scalable Annotation Service — Marken

This post introduces Marken, a new scalable annotation service at Netflix. It details the design goals, including annotating any entity with versioned data, low-latency real-time operations, and offline analytics support. The post explains the schema definition using a JSON-based DSL, geometry objects (WKT), and temporal objects. It also describes the annotation object structure and the use of base schemas for ML algorithms, including multiple inheritance. The architecture is outlined, choosing Cassandra for source of truth and Elasticsearch for search, considering requirements for low latency, horizontal scalability, and high RPS.