Runway - Model Lifecycle Management
summary
Runway is Netflix’s model registry and deployment gate that manages thousands of personalization models, parsing model files, tracking lineage, and enforcing validations before publication. Key components: Scala backend with akka-http, AWS Aurora for relational metadata, Java/Python SDKs, React UI, and a Kafka→Hive→Spark batch pipeline that aggregates consumer load so stillness checks run every 10 minutes. Features include load-validation, single-pipeline publish lock, stillness alerting (example threshold 48h), parent/submodel linking, rollback by redeploying prior instance, and a reported drastic drop in bad-model releases.
AI-written summary · about
chapters (10)
- 0:00Start
- 4:03Discover Models
- 4:50Track Model Instances
- 6:03Model Instance Visualization
- 6:39Model Instance Details
- 7:13Model Monitoring
- 8:11Model Consumers
- 9:23Global View of Unused Models
- 9:53Global View of Alerted Models
- 12:33Architecture
related talks
19mModel Performance Monitoring - A Practitioner's Perspective
40mMachine Learning
25mModel as a Service for Real-Time Decisioning
04mInterview
46mProduction Model Deployment
27mDeveloping and Deploying Models at Scale — without Spark!
04mInterview
31mMLOps In Practice – How To Run Your Machine Learning Models In Production At Enterprise Scale