Transforming Hulu’s Application Platform
summary
Hulu runs Donki, a Heroku-style Docker-native platform that scales from hundreds of requests per second to handling “hundreds of billions of requests per day” for 28 million subscribers. The platform grows from bare-metal Rails/Python deployments to Docker, moves scheduling through Mesos + Aurora then to ECS, and is migrating toward Spinnaker and Kubernetes. Metrics: ~14,000 apps created, ~3,000 apps running, ~83,000 containers, 33 clusters, and about 97% of traffic (≈1.5M req/s) served by Donki. The team prioritizes git-push abstraction, centralized logging, and incremental cluster orchestration.
AI-written summary · about
chapters (17)
- 0:002012 was a great year
- 2:26Service Oriented Architecture
- 2:55Downtime Distractions
- 3:30Desire for faster Time To New Service
- 5:59Focus on commonality: • HTTP server • Load Balancing • Logging
- 7:19Code Requirements: • Health Check URL (for LB) • WSGI module • Logs to stdout
- 12:28Saddle-up
- 13:20An alternative instance
- 14:52Simplify process & threading server scaling knobs
- 15:32File system isolation?
- 16:17Application Memory?
- 17:41System packages?
- 19:24Language & asset precompilation • Artifact promotion
- 22:59Challenges: • Draining nodes for maintenance • Rescheduling nodes on machine failure
- 25:04Challenges: • Autoscaling • Load balancer integration • Cluster spin-up
- 26:10Steps: • Build Docker image • Generate ECS config • Hand off to ECS
- 35:28Challenges: • Custom interface • Artifact promotion • Continuous Deployment via Jenkins is hard
related talks
36mDocker Inside/Out
58mKeynote: So, I have all these Docker containers, now what?
30mHotstar's journey of migrating infrastructure from EC2 to containers
52mContinuous Deployment to the Cloud using Spinnaker
43mDistributed Automation using Kubernetes, Docker, Traefik and Helm
01h 09mDocker in Production
29mHeroku-like Deployments in Your Pocket with Dokku
40mGoes Beyond CI to Production Scale PaaS with Docker