How Replit manages hundreds of thousands of containers
summary
Replit runs hundreds of thousands of Linux containers to give free compute, with ~100-ish containers per VM behind load balancers, websocket connections carrying protobuf messages, cryptographically signed tokens for CPU/memory quotas, and a custom container manager called conman plus a small pid1 process to terminate sessions. Infrastructure splits into control plane / data plane separation and failure domains (free “global” vs paid “hackers”) to localize outages. Abuse detection uses a simple fast model with automated takedowns and human review; regex→substring optimizations save a full CPU core.
AI-written summary · about
chapters (11)
- 0:00Replit introduction
- 1:03Replit mission
- 5:43Rebel spaces and containers
- 7:07Browser compute early
- 10:13Moving to cloud containers
- 15:33Container architecture lessons
- 17:16Failure domains for resilience
- 22:26Control and data separation
- 26:01Free compute security
- 35:38Closing remarks
- 36:00Q&A
related talks
39mHow we learned to love the Data Center Operating System
06mSponsored Keynote: What I Learned Running 10,000+ Kubernetes Clusters
47mCloud Native and Sustainable, Reproducible Scientific Computing
01h 02mContainers in Production: It’s Like Orchestrating Cats
43mContainers Change Everything
38mThree Years of Lessons Running Potentially Malicious Code Inside Containers
40mCompiling to Containers
19mKeynote: Operating 600+ Mesos Servers on 7 Datacenters @Criteo