mcpField
G
gke-compute-class-creator

gke-compute-class-creator

github.com/googlecloudplatformCloud & DevOps16947 installsUpdated 2d ago
Repository

About gke-compute-class-creator

A skill for constructing GKE ComputeClass resources, intended for GKE users who need declarative node configuration and autoscaling behavior. It builds and validates YAML based on cost, performance, or AI/ML requirements and provides an apply command.

Key features

  1. 01Choose a strategy from requirements
  2. 02Set Spot priorities and fallbacks
  3. 03Specify machine family or type
  4. 04Configure GPU or TPU fields
  5. 05Build and validate YAML manifests

Use cases

  1. 01Set Spot fallbacks for cost optimization
  2. 02Choose machine settings for performance
  3. 03Configure node resources for AI/ML
  4. 04Create a ComputeClass and apply it to a cluster

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/googlecloudplatform/gke-mcp --skill gke-compute-class-creator

After installing, type/gke-compute-class-creatorin Claude Code to confirm it loaded.

Bundled filesreferences/compute-class-examples.mdreferences/compute-class-spec.md

Details

From the project README

Creating GKE ComputeClasses

This skill helps you construct ComputeClass resources for Google Kubernetes Engine (GKE). ComputeClasses allow for declarative node configuration and sophisticated autoscaling behaviors like fallback priorities and active migration.

Workflow

  1. Analyze Requirements: Determine the user's goals (Cost optimization? Specific hardware? High availability?).
  2. Select Strategy:
    • Cost Optimization: Use spot: true as a high priority, with spot: false as a fallback.
    • Performance: Select specific machineFamily (e.g., c3, c4) or machineType.
    • AI/ML: Configure gpu or tpu fields.
  3. Construct YAML: Use the references below to build the ComputeClass manifest.
  4. Validate: Ensure all fields comply with the specification.
  5. Apply: Provide the user with the kubectl apply -f <filename>.yaml command.

References

  • Specification: Detailed breakdown of the ComputeClass CRD fields (priorities, machineFamily, gpu, etc.).
  • Examples: Copy-pasteable YAML patterns for common scenarios (Spot fallback, GPU, Zonal).

Usage Tips

  • Active Migration: If the user wants to automatically move back to Spot VMs when they become available, ensure spec.activeMigration.optimizeRulePriority is set to true.
  • Node Selection: Remind the user that to use the class, their Pods must specify: 
    nodeSelector:
  cloud.google.com/compute-class: "<class-name>"
  • Conflict Warning: Advise against mixing ComputeClass selection with other hard node selectors (like cloud.google.com/gke-spot) as this can lead to scheduling conflicts.

Comments

Markdown supported