Control Theory

on YouTube ↗34m

" summary

Travis CI hits a hard cap of ~500 concurrent jobs so queued jobs spike into a red backlog; AWS Auto Scaling only offers metric, schedule or fixed-step policies and new workers take ~20 minutes to accept jobs. The team builds metric-driven and scheduled scaling via the AWS API and experiments with larger headroom (more always-on VMs) and historical prediction (schedule extra instances Monday 09:00 UTC); costs drop queue time but delays still cause surges to miss SLAs.

AI-written summary · about

" chapters (7)

1. 0:00Introduction and welcome
2. 4:37Travis CI origin story
3. 11:15Engineering capacity limits
4. 13:54Control theory basics
5. 25:06Applying control theory to CI
6. 32:23Solutions and ongoing challenges
7. 33:07Closing and Q&A