Ensuring a Consistent Installation Time for ROSA with Hosted Control Planes
summary
Walks through creating thousands of hosted control planes using hcp-burner and workload driver kuerer to measure where installs stall, finds a 10→5 minute timeout bug causing 20–25 minute validating spikes, and fixes it. Implements a cluster-proportional-autoscaler pattern that pre-buffers small "buffer" pods per expected hosted control plane, tightening variance and cutting install median from ~14.5 minutes to ~8:50 — roughly a 6 minute reduction — and demonstrates the method scales and stabilizes installs in multi-tenant environments.
AI-written summary · about
chapters (12)
- 0:00ROSA with hosted control planes
- 2:10Hosted control planes and ROSA
- 7:22Goal of consistent install time
- 8:27Tools for install performance testing
- 9:54ROSA installation flow breakdown
- 12:25Bottlenecks in validating phase
- 14:32Resource scaling challenges
- 16:36Cluster proportional autoscaler
- 19:50Autoscaler results
- 21:57Variable workloads on shared clusters
- 26:52Future improvements
- 27:20Q&A
related talks
27mEfficiency and Innovation in Kubernetes with Hosted Control Planes (HCP)
29mProactive cluster autoscaling in Kubernetes
15mKeeping a Hypervisor Fleet Up to Date with Minimal Customer Disruption
55mTerrible Ideas in Kubernetes
45mLarge-scale kubernetes deployment, which tools to use and how - Rachid Zarouali
34mPanel: Improving and Managing Kubernetes at Scale
37mKubernetesception: How to run a 1000 control planes
36mOptimizing Resource Usage in Kubernetes