Kubernetes Podcast from Google Kubernetes at Uber with Lucy Sweet
22 snips
May 13, 2026 Lucy Sweet, Staff Software Engineer at Uber who led the move of massive compute fleets to Kubernetes. She talks about why Uber migrated from Peloton, making services portable, traffic strategies and canaries. Blue-green node pools, DNS as a critical dependency, stateful migration challenges, and tools like Grail and kubegpt also come up.
AI Snips
Chapters
Transcript
Episode notes
Portability Makes Large Infrastructure Changes Invisible
- Portability enables platform-level changes to be touchless for most teams, like shifting CPU architectures.
- Uber switched workloads to ARM by trialing containers under make-before-break and only contacting owners for persistent failures.
Use Small Canary Traffic Budgets For Safe Rollouts
- Do use canaries (e.g., 1% traffic) during rollouts to detect regressions before full cutover.
- Uber routes ~1% of traffic to canaries and provides customizable options, keeping platform defaults simple for developers.
Node Readiness Should Be Workload Aware
- Node readiness can be workload-dependent; a binary node-ready signal is often insufficient.
- Lucy argues for expressing node readiness per workload type to handle GPU/accelerator placement and nuanced upgrades.