Skip to content
This repository has been archived by the owner on Aug 9, 2024. It is now read-only.

Latest commit

 

History

History
67 lines (39 loc) · 3.12 KB

README.md

File metadata and controls

67 lines (39 loc) · 3.12 KB
Raw

Kubeflow Community Infrastructure

This is a repository for using declarative configs and GitOps to managed shared community Kubeflow GCP infrastructure.

The management cluster is setup using the Kubeflow management blueprint.

Creating GCP Resources.

If you need to create GCP resources for Kubeflow or gain access to GCP resources you do so by creating PRs against this repository.

  • We use ACM to sync the Cloud Config Connector(CNRM) to GKE cluster that will apply those resouces.

  • ACM has an oppinionated layout to the repository which is rooted at "/prod"

    • See the docs for how this repository should be layed out

    • There should be a namespace for every GCP project that is managed

  • Follow these steps to create new project. Note that ${PROJECT} name must be globally unique across all GCP projects.

    1. Create subfolder /prod/namespaces/${PROJECT}.

    2. Create /prod/namespaces/${PROJECT}/namespace.yaml defining a Kubernetes namespace. Namespace name should be equal to ${PROJECT} name.

    3. Create /prod/namespaces/${PROJECT}/project.yaml containing a Project resource defining your project.

    4. Create /prod/namespaces/${PROJECT}/iam-policy-members.yaml containing a IAMPolicyMember resource list with necessary IAM permissions to access the project. Each IAMPolicyMember should have unique name.

      You can give roles/editor to your GCP user account to view created project.

      If you want to integrate your project with kubeflow-ci, you have to give access to this service account: serviceAccount:kubeflow-testing@kubeflow-ci.iam.gserviceaccount.com.

      kubeflow-testing service account should have these permissions:

      • roles/editor to modify GCP resources.
      • roles/cloudbuild.builds.editor to create Cloud Builds.
      • roles/container.admin to manage Kubernetes clusters.

  • Wait for the PR to be approved

  • Once the PR is merged the resources should be created automatically and you can access created GCP project. You can run kubectl describe on appropriate resource in kf-community-admin cluster to check status.

Setup

  1. Follow the management blueprint

    • Do not install CNRM; we will use ConfigSync to install CNRM

  2. Follow the ACM installation guide

    • Create the service account 'cnrm-system' in project kf-kcc-admin
    • Note It looks like when using ACM to install and manage CNRM you can't use workload identity and need to provide a GCP service account key.

  3. Make sure the CNRM service account has roles roles/owner and project creator on the community folder