Reliably shipping containers in a resource rich world using Titan
summary
Netflix runs Titan to provision and supervise thousands of containers across three AWS regions, combining Docker and Apache Mesos with a custom scheduler, ZFS-backed volumes for snapshots and movement, and a logging API that streams container logs to S3. Titan ties container lifecycle to process supervision, supports cross-region clusters, offers both reactive and predictive autoscaling (Scryer linear regression), and experiments with a Docker network plugin. It already runs mission‑critical batch jobs and plans to open source the project.
AI-written summary · about
chapters (13)
- 0:00Introduction to Titan
- 1:16Titan core concepts
- 4:03Netflix infrastructure context
- 7:14Designing for developers
- 10:48Operations and efficiency
- 14:53Choosing Docker
- 18:05Architecture overview
- 20:14System components
- 22:45Logging and networking
- 25:41Monitoring and fault tolerance
- 28:35Auto-scaling strategies
- 30:44Roadmap and future outlook
- 32:38Q&A session
related talks
58mDistributed Scheduling with Apache Mesos in the Cloud
50mFast Delivery
39mPractical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at Netflix
27mBreaking the monolith with Node and Docker at Netflix
42mRunning a Massively Parallel Self-Serve Distributed Data System at Scale
01h 04mLaunching patterns for containers - it's more than just scheduling
19mKeynote: Making and Keeping Netflix Highly Available
42mLive Coding with Containers