Learnings & Patterns from Zero-Downtime Migrations
summary
Conductor migrates its primary persistence from Redis-based Dynamite to Cassandra + AWS S3, moves queues from Dyno queues into a new Timestone system, and uses a migration layer with a switch pattern and dual-write/read logic to route new traffic while allowing in-flight workflows to finish. The rollout produces no downtime, drains Dyno queues, and relies on tens of full-scale test runs; about 0.3% of workflows required opportunistic read-before-write handling.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:58Agenda
- 1:39Netflix Studio
- 2:49Catalysts
- 3:36Constraints
- 4:00Conductor
- 4:40Conductor Architecture
- 5:11Case Study 1
- 5:22Triggers
- 6:33Storage Issues
- 7:42Persistence Layer
- 9:10Migration Diagram
- 10:40Case Study
- 12:39Migration Layer
- 13:55Test Setup
- 15:34Patterns
- 17:22Learnings
related talks
34mKilling the Dinosaur: A migration story
01h 01mLegacy Architecture Migration Patterns with DDD
50mMigrating Your Most Critical Feature Without Regressions or Downtime
32mChanging Engines Inflight - A Real Life Modernisation Story
43mLegacy Architecture Migration Patterns
40mModernizing the Legacy
29mTaming the Hybrid Beast with DevOps
32mQuietly Strangling the Monolith - a real-world migration of a high traffic monolith