zuodexin/ais-k8s

Kubernetes Operator, helm charts, and production scripts for large-scale AIStore deployments on Kubernetes.

AIStore on Kubernetes

The repository contains tools and supporting materials for deploying AIStore on Kubernetes.

This includes:

• Ansible playbooks to assist in preparing nodes to host AIStore
• Kubernetes operator
• Documentation, and in particular:
  • Kubernetes Operator Deployment: Steps
• Terraform definitions for public cloud usage, such as GKE/GCP.

Cloud Deployment

If you want to deploy a fresh Kubernetes cluster in the cloud with AIStore, please refer to the terraform directory of this repository.

Small Scale Experimental Deployments

It is assumed you want to deploy AIStore at reasonable scale on multiple nodes each with multiple drives. If you don't require such scale then consider deploying under Docker as illustrated in the main AIStore repo.

Deployment Documentation

You can deploy AIStore on Kubernetes in two ways. In both cases, some preparation and planning is needed; we suggest you read the deployment documentation first.

Deployment via the AIStore Operator

AIStore is deployed using the AIStore operator.

With an operator based deployment, instead of deploying services directly, you define your AIStore cluster as a kubernetes custom resource.

The operator documentation can be found here, along with detailed walkthrough guidance.

Using This Repository For GitOps-Style Deployment

We suggest cloning this repository and retaining the master branch as tracking this upstream master; create a new branch off of master and edit values.yaml etc., and point your CD tool at that branch. When you pull updates to the master you can pull and merge them into your private branch.