[FEATURES] Introduce Fluid CLI Tool for Inspection and AI-Assisted Diagnosis

[cheyang]

What feature you'd like to add:
Fluid manages Kubernetes resources—such as StatefulSets, PersistentVolumeClaims, and PersistentVolumes—through two custom resources: Dataset and Runtime. While powerful, debugging issues or understanding the current state of these underlying resources can be challenging for users, often requiring manual inspection across multiple Kubernetes objects.

To improve developer experience and operational efficiency, we propose a dedicated CLI tool (ideally implemented as a kubectl plugin) that provides intuitive commands to inspect and diagnose Fluid-managed workloads. What we need are:

• Design and implement a CLI tool named kubectl-fluid (or similar) as a kubectl plugin.
• Implement an inspect subcommand that, given a Dataset name and namespace, lists all associated Kubernetes resources (e.g., Runtime pods, PVCs, PVs, services) along with their statuses.
• Implement a diagnose subcommand that:
  • Collects relevant diagnostic data (e.g., pod logs, events, resource states, controller conditions).
  • Packages this information into a structured summary for troubleshooting.
• Build a lightweight framework to support AI/LLM-assisted diagnosis:
  • Format collected diagnostic context into a prompt-ready structure.
  • Provide an optional integration path (e.g., via config or flag) to send context to an LLM endpoint for analysis (future extensibility; no external dependency required at build time).

Why is this feature needed:
This tool will significantly lower the barrier to debugging Fluid deployments, especially for new users or in complex failure scenarios. The inclusion of an AI-ready diagnostic framework positions Fluid at the forefront of intelligent cloud-native observability.

[dikshant182004]

hi @cheyang @TrafalgarZZZ , i would like to contribute on this project .I do have a decent experience with LLM ,data gathering and ETL flows which is needed for the CLI tool .Is there any pre test or contribution criteria to get selected for this project or we just have to apply at the LFX portal with a good proposal defining our approach and how we can complete this task + can u provide the communication channel link as slack channel doesn't seems present or active

One more thing i want to ask is that can't we add some vizualization stuff or make a dashboard or something here as we are vizualizing the kubernetes resources like we do in cartography for cloud resources and it would give more user friendly appearance to display the resources workloads using some distribution and graphs rather than just providing the numerical figures using CLI and we can make the dashboard as a one stop place to see the numerical data with figures and vizualization along with some AI inference.

[devvaansh]

Hi @cheyang
I’ve been researching the Fluid landscape and would love to contribute to kubectl-fluid.
While generic tools like kubectl-tree follow basic ownership, they miss the "sideways" dependencies critical to Fluid, like linking a Dataset to its CSI Mount Pods and the application's data affinity. I want to implement a structural data-path traversal that maps these implicit relationships, giving SREs a true "topology of data" rather than just a tree of resources.
I’m also keen on integrating a PII-sanitized AI diagnostic flow, ensuring logs are redacted locally before being passed to an LLM provider.

[samruddhi-Rahegaonkar]

Hi @cheyang ,
I’m interested in working on this feature as part of LFX Mentorship 2026 Term 1.
I plan to start with a minimal kubectl plugin scaffold and an initial inspect command before extending to diagnostics.
I’d appreciate any guidance on prioritizing scope or preferred design patterns.
Thanks!

[kryptonn36]

Hi @cheyang ,

I’m interested in contributing to the kubectl-fluid CLI for LFX 2026 Term 1. I have solid experience with Go and building CLI tools but am new to Kubernetes. I’m eager to learn and contribute, especially on the inspect and diagnose commands and the AI-assisted diagnosis part.

Any guidance on scope or design would be appreciated. Looking forward to contributing!

[Dei405]

Hi @cheyang

I'm very interested in this feature; I have some experience with Go and build kubectl plugins.
I want to learn how to extend the CLI to help LLM diagnose problems.

Thanks!