Teaching Kubernetes to Scale with a MacBook Screen Lock, with Brian Donelan
Aug 26, 2025
Brian Donelan, VP of Cloud Platform Engineering at JPMorgan Chase, shares his innovative side project that automates Kubernetes workload scaling based on MacBook screen lock status. He connects macOS notifications to CloudWatch, achieving impressive 80% cost savings by scaling resources to zero when idle. The discussion highlights KEDA's unique event-driven scaling capabilities, creative metrics for different industries, and strategies for optimizing cloud resource usage, making workload management more efficient and sustainable.
AI Snips
Chapters
Transcript
Episode notes
Side Project: MacBook-Controlled Cloud Services
- Brian ran resource-heavy dev services in the cloud because his MacBook couldn't handle them.
- He wanted an automated way to turn them off when he closed his laptop to avoid surprise bills.
Rethink Scaling Signals Beyond CPU/Memory
- CPU and memory are poor proxies for many event-driven workloads and react slowly to demand.
- Using leading indicators can create much more elastic and cost-effective scaling.
How Screen Lock Triggers Scale-To-Zero
- A macOS daemon ships a boolean metric (0 locked, 1 unlocked) to CloudWatch every minute.
- KEDA polls that metric and scales pods to zero or one, while Karpenter consolidates nodes accordingly.