Hello everyone, we are currently working on utilizing LLM agents for autonomous management of the Kubernetes cluster. We have published a study on this in arXiv([2407.14402] The Vision of Autonomic Computing: Can LLMs Make It a Reality?).
We are interested in collaborating with Kubernetes experts to delve deeper into this research.
We welcome messages from the community and are excited to work together to make a positive impact on Kubernetes ecosystem!
I’d argue that LLM won’t help improve k8s cluster management despite it may make things look fancy. However, there are other AI techniques which are possibly applicable.
Thanks for your comments! What are the other techniques you think will be more helpful?
For example, Causal AI, pattern recognition, etc.
Cluster management doesn’t always mean that we need to deal with a huge amount of unstructured data. We may have a lot of log data or metrics data, but I’m not optimistic that LLM is smart enough to perform any useful root cause analysis work. In this narrow field, what we really need is the insight into the way a cluster is running/failing, and a tool suite that captures this insight. Even if we can come up with a user-friendly Q/A system with LLM, the system would be still difficult to use for a user who knows nothing about Kubernetes, and it would be too naive or misleading to an expert. Maybe I’m not quite clear about the value proposition.