Building a Flexible, Scalable Infrastructure to Test Dolby.IO at Scale
summary
A Kubernetes cluster using Kubernetes + AWS Carpenter runs a scale-testing platform that accepts a DSL YAML blueprint to declare scenarios, spins heterogeneous clients (real web/C++ SDKs and thin aiortc-based clients) via forked locust job specs, and coordinates them with Redis pub/sub. Current baseline supports 50 active participants and 100k listeners; goal is 1 million listeners and 25 presenters. Costs steer spot versus on‑demand instances, single-job queuing avoids interference, and pod restart skew remains a key unresolved failure mode.
AI-written summary · about
chapters (10)
- 0:00Scalable test infrastructure
- 1:29Conferencing capacity goals
- 4:00Testing goals and KPIs
- 9:36Limitations of tools
- 11:43New architecture overview
- 13:49Client workflows
- 16:19Publish-subscribe model and resource handling
- 17:46Test scenario DSL
- 20:40Future improvements
- 21:47Summary and open floor for questions
related talks
42mThe Challenges Managing a Kubernetes-Based Machine Learning Infrastructure
31mLoad Testing Kubernetes: How to Optimize Your Cluster Resource Allocation in Production [I]
35mIntro: Testing SIG
45mBuilding loose-coupled continuous testing platform
01h 08mPerformance testing in the cloud
29mHow to Build a Cluster to Run Hundreds of Thousands of High-CPU/High-MEM Jobs
23mDistributed performance testing at the push of a button
42m18 Months Before the Mast