AI/ML Project Demo - k8s-ai-autoscaling

Contributor

Name: [Orkhan Mahmudzada]
Email/Slack: [[email protected]]

---

Project Overview

Project Name

[k8s-ai-autoscaling]

Project Purpose

[AI-Based Forecasting and Autoscaling with Prophet,Prometheus and KEDA.]

Project Outcomes

• [Accurate Forecasting of Resource Usage]
• [Improved Autoscaling Strategy]
• [Scalable and Flexible Architecture]

---

Technical Details

Platform

• Runs on: k0rdent

Repo

• Hosted at: [https://github.com/Mirantis/k8s-ai-autoscaling]

Technical Overview

• Uses Facebook Prophet for time-series forecasting of resource metrics.
• Metrics collected via Prometheus from Kubernetes workloads.
• Forecasts exposed via a REST API, scraped by Prometheus using a ServiceMonitor.
• KEDA scales deployments based on forecasted metrics.
• Deployed via Helm, with support for namespace isolation and dependency management.

---

Service Templates

Service Templates Used

[KEDA]

Service Templates Created Specifically for this Project

[ai-metric-analyzer-0.1.0]

---

Documentation and Links

• 📄 Design Document
• 🎥 Demo Video
• 📘 GitHub Repo

---

Artifacts

• Docker Image: python:3.10-slim
• Helm Chart: ai-metrics-analyzer/ (under Mirantis repo)
• Custom ServiceMonitor definition

graph TD
    Prometheus -->|scrapes| AI_Metrics_Analyzer
    AI_Metrics_Analyzer --> Prophet
    AI_Metrics_Analyzer -->|exposes| /metrics
    KEDA -->|reads| Prometheus
    KEDA -->|scales| Target_App

Loading

---