Forget Kubectl and Talk to Your Clusters: Using LLMs to Simplify Kubernetes Cluster...
Talkat KubeCon + CloudNativeCon + Open Source Summit China 2023 · on Thursday, 2023-09-28· at 13:55· 3层 305B会议室| 3F Room 305B·
summary
A prototype converts natural-language intents into deterministic API calls by fine-tuning local LLMs on 40,000 example pairs to extract entities and map to actions (initial models include internal GB/D variants), replaces basic "kubectl get" workflows, and adds a SQL cache layer that turns watched Kubernetes resources into queryable tables plus a federation layer across clusters. Results: high NER and intent accuracy but persistent hallucination and long-context limits (≈4,000-token requests); deployed as read‑only with human verification.
AI-written summary · about
chapters (10)
- 0:00LLMs for Kubernetes
- 2:50Motivations for LLM approach
- 6:23Goals and requirements
- 10:39First API experiment
- 16:15User constraints challenges
- 20:49LLM YAML testing
- 23:57SQL and caching solution
- 26:06Achievements and limitations
- 29:00Lessons learned and future
- 31:29Q&A
related talks
31mCluster Operations as a Service, Introducing LLM Backed Controllers.
33mScalable LLM Inference on Kubernetes With NVIDIA NIMS, LangChain, Milvus and Flu...
07mLightning Talk: Demystifying Kubernetes Observability with Generative AI and LLMs
14mReasoning Operators: Bringing LLM Logic Into Kubernetes Control Loops
30mSecuring LLMs in Kubernetes: Best Practices
09mTransform Your Kubernetes Cluster Into a GenAI Platform: Get Ready-t...
08mKeynote: Plug in and Scale: Serving LLM Models on Kubernetes Ma...
33mBeyond VLLM: Distributed LLM Inferencing With Llm-d on Kubernetes