The Karpenter Effect: Redefining Kubernetes Operations, with Tanat Lokejaroenlarb
11 snips
Nov 18, 2025 Tanat Lokejaroenlarb, an SRE at Adevinta leading the runtime team for the Ship platform, dives into his journey migrating to AWS Karpenter. He reveals how replacing EKS Managed Node Groups enhanced Kubernetes operations, cutting costs by €30,000 monthly. Tanat discusses innovative strategies like decoupling upgrades, implementing automated instance selection, and using Kyverno for policy automation. Learn about over-provisioning with low-priority pods and the significant performance benefits of AMD adoption. His insights are a must for anyone in cloud-native operations!
AI Snips
Chapters
Transcript
Episode notes
Migration From MNGs To Karpenter
- Tanat replaced EKS Managed Node Groups and Cluster Autoscaler with AWS Karpenter across 30+ clusters and ~2,500 nodes.
- The migration solved brittle upgrades and instance flexibility problems at scale.
Decoupling Control Plane And Nodes
- Karpenter decouples control plane upgrades from node rollouts, making upgrades asynchronous and less risky.
- The control plane upgrade takes ~15–20 minutes while Karpenter handles node rollout in the background with pacing control.
Respecting Pod Disruption Budgets
- Karpenter respects Pod Disruption Budgets (PDBs) and waits for safe eviction, unlike their old autoscaler that forcibly terminated nodes.
- They supplement PDBs with Karpenter Node Disruption Budgets to control disruption pace (e.g., 1–3 nodes per 15 minutes).