4.0 KiB
title, type, source-type, category, tags, date-added, video-source, audio-source, status
| title | type | source-type | category | tags | date-added | video-source | audio-source | status | |||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Public Cloud Learning Sessions - EKS Optimization part 3 of 3 - Introduction to EKS Auto Mode - 20250304 170115-Meeting Recording | cloud-learning | video | DevOps & SRE/04_EKS |
|
2026-04-14 | nas:///volume2/work/Public Cloud Learning Sessions/Public Cloud Learning Sessions - EKS Optimization part 3 of 3 - Introduction to EKS Auto Mode - 20250304_170115-Meeting Recording.mp4 | summarized (Gemini 摘要) |
Public Cloud Learning Sessions - EKS Optimization part 3 of 3 - Introduction to EKS Auto Mode - 20250304 170115-Meeting Recording
Source: NAS /volume2/work/Public Cloud Learning Sessions/Public Cloud Learning Sessions - EKS Optimization part 3 of 3 - Introduction to EKS Auto Mode - 20250304_170115-Meeting Recording.mp4
Type: VIDEO | Category: 04_EKS
Status: 🟡 Awaiting Whisper transcription → Summary
EKS Optimization: Introduction to EKS Auto Mode
This session focuses on EKS Auto Mode, the third part of a series on EKS optimization. EKS Auto Mode extends the management responsibilities of the EKS service to the data plane, managing instances, operating systems, patches, and security updates. It leverages core capabilities like Carpenter for infrastructure management, a managed EBS CSI driver for stateful workloads, and the AWS load balancer controller.
Key benefits of EKS Auto Mode include increased agility, automatic consolidation, dynamic instance determination, and optimized compute costs. With Auto Mode, a majority of the operational concerns are being managed by the ECS service. Core capabilities are managed within instances provisioned inside the EKS account, while customers retain control over VPC infrastructure, cluster configuration, add-ons, and workload configurations.
EKS Auto Mode offers an easier interface for working with EKS, providing data plane management in addition to control plane management. It supports a wide range of EC2 instances (excluding bare metal) and is fully compatible with Kubernetes-compliant workloads. Security is enhanced through the use of the Bottle Rocket operating system and automated patch management. The core cluster capabilities are grouped under compute (Carpenter controller), networking (AWS load balancer controller), storage (EBS CSI controller), and security (pod identity associations).
By default, Auto Mode includes two node pools (general purpose and system) and one node class. The default node pools are immutable and configured with zero weight, allowing custom node pools to be prioritized. The general purpose node pool is locked to AMD64 architecture, while custom node pools can be defined for Graviton instances. Instances in the system node pool have a taint applied, requiring corresponding tolerations for system add-ons.
Networking in Auto Mode includes Core DNS packaged with every node as a system service, VPCCNI as a system service, and Kube proxy set up in IP tables mode. Prefix delegation is enabled by default. The AWS load balancer controller is available as a core capability, using an EKS Auto Mode-specific load balancer class. The packaged CSI controller requires a storage class referring to the EBS CSI EKS provisioner.
Version upgrades in Auto Mode are initiated by an operator for the control plane. Once the control plane version gets upgraded, then the compute controller, which is running as a core capability, will identify that the control plane version has changed and it will try to pull the current AMI version for that new control plane version. The compute controller then rolls out the new AMI across the cluster through a rolling upgrade.
While the controllers are managed by the EKS service, users can investigate custom resources and deploy node diagnostic CRDs. Observability can be achieved through CloudWatch agent, AWS distro for open telemetry, or other collectors.
For every instance spun up in an Auto Mode cluster, there is a 12% premium charged for the automatic management of those instances.