Keynote: Making and Keeping Netflix Highly Available
summary
Netflix runs ~380 streaming-path microservices for 100+ million customers across ~1,000 device types and uses a Mesos-based stream processor Mantis plus a cloud-optimized scheduler Fenzel on AWS EC2 to get operational insights. 125 million viewing hours per day drives spikes from ~6M to ~18M messages through Mantis, and a real-time SPS pipeline detects playback drops in seconds. Teams create on-demand Mantis jobs for targeted debugging; auto-scaling clusters and regional traffic shifts provide rapid mitigation and reduced user impact.
AI-written summary · about
chapters (7)
- 0:00Intro
- 1:54By the numbers.
- 5:41Mantis overview
- 7:31Core architectural components
- 9:15Real-time SPS
- 12:37Real-time metrics
- 15:12Ad-hoc queries
related talks
27mBehind the Scenes: Running Critical Services @ Netflix
31mKeeping Your Cloud Footprint in Check
49mNetflix Keystone - Real-time Stream Processing Cloud Service
44mTechniques Netflix Uses to Weather Significant Demand Shifts
39mPractical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at Netflix
34mBeaming Flink to the Cloud @ Netflix
46mA Day in the Life of a Netflix Performance Architect
27mKeynote Panel: Future of Cluster Management