Google Cloud believes it has the solution to maximising the usage of Kubernetes (an open-source container orchestration system for automating software deployment, scaling, and management) within organisation, resulting in cost savings and increased productivity.
In an effort to inform customers of the entire range of capabilities of the container system and how to maximise efficiency without sacrificing the end-user experience or the dependability of related apps, the cloud computing service has produced a study on how to run clusters of the container system.
Some of the findings of the report include the significance of establishing suitable resource requests, the difficulty some clusters have in striking a balance between cost and efficiency, and how top performers take advantage of cloud discounts.
Authors Anthony Bushong, Developer Relations Engineer at Google, and Ameenah Burhan, Solutions Architect at Google, claim to have conducted a “large-scale analysis of Kubernetes clusters” and have discovered four “golden signals” for cost optimisation while preserving workload stability.
Data from Google Kubernetes Engine (GKE) clusters was anonymized and sorted based on how well they performed in comparison to the signals.
The survey discovered that many users aren’t setting requests for their workloads, despite the fact that this is the most crucial thing to do. This is an issue, according to the authors, because “Kubernetes reclaims resources when node-pressure occurs.” Requests must still be set for workloads that call for a minimal level of reliability.
Requests are assigned to Pods under the Best Effort Quality of Service (QoS) class if no requests are specified. These are the workloads that are most likely to be terminated if resources are limited on a particular node, which might cause uneven performance and reliability problems. Furthermore, when such problems arise, they might be challenging to debug.
Bushong says, fortunately, both a script using the kube-requests-checker and the GKE Workloads at Risk dashboard makes it simple to find workloads without defined requests.