Enhancing Kubernetes Cluster Troubleshooting Efficiency With K8sGPT
Puutio, Kalle (2025)
2025
Tietojenkäsittelyopin maisteriohjelma - Master's Programme in Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-06-13
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202505286297
https://urn.fi/URN:NBN:fi:tuni-202505286297
Tiivistelmä
As AI-assisted workflows gain more prominence, exploring new applications of AI in various contexts has become essential. This thesis explores how users interact with and use AI-powered Kubernetes triaging too K8sGPT to diagnose, debug and fix errors in Kubernetes cluster.
The study aims to analyze user’s experiences using K8sGPT to debug and fix issues in Kubernetes cluster. The main research questions are “How can K8sGPT enable users to debug Kubernetes clusters more efficiently?” and “How do users utilize K8sGPT?”. The data was collected using Thematic Interview after the user completed a series of Kubernetes tasks during a one hour period, and the results analyzed using Reflexive Thematic Analysis.
The findings indicate that K8sGPT’s AI-powered overview of the cluster state, due to being concise and human-readable, reduces the user’s manual workload and mental load, and enables rapid onboarding into the debugging process. The AI-generated suggested solutions enable beginner- and occasional-level users to solve Kubernetes issues they would not be able to solve otherwise. Due to readability issues in K8sGPT’s output and lack of domain knowledge, less skilled users rarely formed a comprehensive picture of the issue at hand and attempted to fix the symptoms instead of identifying and fixing the root cause.
While users rely on K8sGPT throughout the triaging process, they also use another AI to understand K8sGPT’s output. The lack of domain knowledge can lead to overreliance on AI, which in turn lead to the inability for users to fix the issues in cluster. K8sGPT and similar AI-powered tools should be elaborative enough to allow users to understand the output comprehensively, and allow the users to ask the AI to elaborate more in case the users require more explanation. Users should have the ability to customize the interface and output to their liking, and normal efficiency features such as shortcuts and quick keys should be implemented.
The study aims to analyze user’s experiences using K8sGPT to debug and fix issues in Kubernetes cluster. The main research questions are “How can K8sGPT enable users to debug Kubernetes clusters more efficiently?” and “How do users utilize K8sGPT?”. The data was collected using Thematic Interview after the user completed a series of Kubernetes tasks during a one hour period, and the results analyzed using Reflexive Thematic Analysis.
The findings indicate that K8sGPT’s AI-powered overview of the cluster state, due to being concise and human-readable, reduces the user’s manual workload and mental load, and enables rapid onboarding into the debugging process. The AI-generated suggested solutions enable beginner- and occasional-level users to solve Kubernetes issues they would not be able to solve otherwise. Due to readability issues in K8sGPT’s output and lack of domain knowledge, less skilled users rarely formed a comprehensive picture of the issue at hand and attempted to fix the symptoms instead of identifying and fixing the root cause.
While users rely on K8sGPT throughout the triaging process, they also use another AI to understand K8sGPT’s output. The lack of domain knowledge can lead to overreliance on AI, which in turn lead to the inability for users to fix the issues in cluster. K8sGPT and similar AI-powered tools should be elaborative enough to allow users to understand the output comprehensively, and allow the users to ask the AI to elaborate more in case the users require more explanation. Users should have the ability to customize the interface and output to their liking, and normal efficiency features such as shortcuts and quick keys should be implemented.