AI Agent Fixes OOMKilled in Kubernetes 🔥 Intelligent Memory Healing Demo
GitHub Repo:
https://github.com/happy-jitesh/k8s-o...
Repo link: https://tinyurl.com/4w4zb3t5
OOMKilled incidents are NOT restart problems.
In this video, my AI Agent intelligently heals Kubernetes memory failures.
In this episode, we upgrade our Kubernetes Incident Resolution AI Agent
to handle a very realistic production failure:
👉 OOMKilled (Out Of Memory) incidents.
Unlike CrashLoopBackOff failures, restarting pods does NOT fix OOMKilled.
These incidents require resource tuning.
🔥 What changed in this video:
✅ Upgraded from Script → Controller-based AI Agent
✅ Replaced Phi → Llama3 (LLM Brain)
✅ Implemented Intelligent OOMKilled Healing Logic
🧠 AI Agent Behavior:
• Continuously monitors Kubernetes cluster
• Detects OOMKilled pods
• Collects memory limits & restart history
• Uses Llama3 (via Ollama) for reasoning
• Increases memory limits automatically
• Triggers safe rolling restart
• Verifies deployment health
This is NOT basic automation.
This is Agentic AI — autonomous decision-making for DevOps.
🛠️ Technologies Used:
• Kubernetes / Minikube
• Python AI Agent
• Ollama (Local LLM)
• Llama3 Model
• kubectl
📌 Key Learning:
Not all Kubernetes failures should be solved with restarts.
🚀 Next Episodes:
• CPU Throttling Healing
• Probe Failure Intelligence
• Incident Memory System
• Multi-agent DevOps AI
👍 Like if this helped
💬 Comment which DevOps problem AI should solve next
🔔 Subscribe for real Agentic AI for DevOps projects