Скачать или смотреть Understanding Kubernetes Cluster-Autoscaler and Its CPU Utilization Metrics in EKS

Explore how Kubernetes Cluster-Autoscaler determines CPU utilization and the implications for node scaling in Amazon EKS.
---
This video is based on the question https://stackoverflow.com/q/73238459/ asked by the user 'Avanth Aditya' ( https://stackoverflow.com/u/14202646/ ) and on the answer https://stackoverflow.com/a/73537639/ provided by the user 'Dennis van de Hoef - Xiotin' ( https://stackoverflow.com/u/5600652/ ) at 'Stack Overflow' website. Thanks to these great users and Stackexchange community for their contributions.

Visit these links for original content and any more details, such as alternate solutions, latest updates/developments on topic, comments, revision history etc. For example, the original title of the Question was: how is kubernetes cluster-autoscaler determining cpu utilization of nodes in EKS

Also, Content (except music) licensed under CC BY-SA https://meta.stackexchange.com/help/l...
The original Question post is licensed under the 'CC BY-SA 4.0' ( https://creativecommons.org/licenses/... ) license, and the original Answer post is licensed under the 'CC BY-SA 4.0' ( https://creativecommons.org/licenses/... ) license.

If anything seems off to you, please feel free to write me at vlogize [AT] gmail [DOT] com.
---
Understanding Kubernetes Cluster-Autoscaler and Its CPU Utilization Metrics in EKS

When managing Kubernetes clusters, especially in cloud environments like Amazon EKS (Elastic Kubernetes Service), effective resource management is essential for optimal performance and cost-saving. One of the key components in achieving this is the Cluster-Autoscaler, which adjusts the number of nodes in your cluster based on the resource demands of your applications. However, users often face confusion regarding how the Cluster-Autoscaler determines metrics like CPU utilization for scaling nodes.

In this guide, we will dissect a specific example to illustrate how the Cluster-Autoscaler calculates CPU utilization and why nodes may not scale down even when they seem idle.

The Problem: Understanding CPU Utilization Reporting

A user reported an instance where, despite their Amazon EKS cluster showing a node utilizing only 5% of its CPU (and 21% of memory) when checked with kubectl top nodes, the Cluster-Autoscaler logs displayed a message indicating that the node was not eligible for removal due to high CPU utilization (0.663130 or 66.31%).

Key Points of Confusion:

Discrepancy Between Observations: The actual CPU usage appears low, while the autoscaler sees it as high.

Scale-Down Threshold: With a default configuration of --scale-down-utilization-threshold=0.5, the user expected the Cluster-Autoscaler to scale down but is uncertain why it didn't.

The Solution: How Cluster-Autoscaler Calculates CPU Utilization

1. Understanding Requested vs. Actual CPU Usage

The Cluster-Autoscaler primarily evaluates the requested resources of the pods running on a node rather than the actual resource usage. This distinction is crucial:

Requested CPU: The amount of CPU that a pod/container requests from the Kubernetes scheduler. This value is guaranteed, meaning that even if the actual CPU usage is much lower, the requested amount is what matters from the scale perspective.

Actual CPU Usage: This metric reflects the real-time resource consumption of the node.

In the user's case, the CPU utilization of 66.31% correlates more closely with the CPU requested by the pods currently running on that node, rather than the low actual usage reported (5%).

2. Why the Node Is Not Scaled Down

The Cluster-Autoscaler does not use actual CPU usage as its primary decision factor for scaling down nodes. Instead, it looks at the requested resources:

Since the total requested CPU for the pods exceeds the scale-down threshold set, the autoscaler maintains the node.

Even though the node is not heavily utilized, it's still “busy” from the perspective of resource requests.

Conclusion: A Better Understanding of Cluster-Autoscaling

Understanding how the Kubernetes Cluster-Autoscaler calculates metrics like CPU utilization can help users manage their resources more effectively. Here are some takeaway points:

The requested resources of pods take precedence over actual usage for scaling decisions.

Configure your autoscaling parameters based on your application's resource request patterns to maximize efficiency.

By getting to grips with these details, you can better anticipate how Kubernetes will manage your scaling needs and take appropriate steps to ensure your cluster operates smoothly and cost-effectively.

Final Thoughts

If you are managing an EKS cluster, remember that perceived low CPU utilization does not always translate to suitability for scaling down. Ensure your application's resource requests align with your scaling strategy to maintain efficiency and performance.