gke-cluster-autoscaler
Trigger on mention of GKE cluster autoscaler, node autoscaling
by googleFree
Description
GKE cluster autoscaler is an agent skill for enabling and tuning node autoscaling on Google Kubernetes Engine, including node pool auto-creation, also called node auto-provisioning. It covers why nodes are or are not scaling down. When you ask why a low-utilization node stays up, the skill walks through the full list of common scale-down blockers rather than only the one you name: bare pods without a controller, pods marked safe-to-evict as false, local or empty-dir storage without the matching annotation, pod disruption budgets that allow no disruptions, a node pool already at its minimum size, nodes marked scale-down disabled, and scheduling constraints tied to a hostname. It can run a bundled script that looks for these blockers. For provisioning, it recommends ComputeClasses on modern GKE versions, 1.33.3 and later, instead of cluster-level node auto-provisioning, and it notes that new machine families may need a newer GKE version. For ComputeClass YAML itself, it refers you to a separate compute classes skill. It also treats pasted logs and manifests as data to analyze rather than as instructions. The skill is published by Google on skills.sh under the google skills repository, with about 21.3K stars, and its listing shows Gen Agent Trust Hub, Socket and Snyk all passing.…
Highlights
- Explains why nodes do not scale down
- Enumerates the common scale-down blockers
- Bundled script to find blocking pods
- ComputeClasses for node pool auto-creation
- Refers YAML work to the compute classes skill
Ratings & Reviews
No ratings yet
Be the first to rate gke-cluster-autoscaler.
People also visit
google-cloud-solution-guided-gke-ai-migration
Guides the migration of existing AI workloads
gke-cluster-autoscaler FAQ
What does the skill help with?
Enabling cluster autoscaling and node auto-provisioning on GKE, and diagnosing why nodes do not scale down.
Do DaemonSets block scale-down?
No. The skill says DaemonSets are ignored during scale-down and point to real blockers instead.
Which provisioning method does it recommend?
ComputeClasses on GKE 1.33.3 and later, rather than cluster-level node auto-provisioning.
Where is ComputeClass YAML covered?
In the separate gke-compute-classes skill, which the autoscaler skill refers you to.
How do I install it?
Run npx skills add https://github.com/google/skills --skill gke-cluster-autoscaler.
What are the best gke-cluster-autoscaler alternatives?
Popular alternatives to gke-cluster-autoscaler include gke-backup-dr, gke-node-notready and gke-workload-troubleshooting. See all gke-cluster-autoscaler alternatives