GKE Cluster Autoscaler Guide
Configure, optimize, and troubleshoot automatic GKE node scaling.
The document explicitly rejects injected identifiers, treats pasted logs and YAML as untrusted data, and enforces a restricted naming pattern, improving input safety and data-flow clarity. Points are deducted because broad user confirmation, pre/post-change checks, rollback procedures, and least-privilege guidance are incomplete. It also recommends granting the Editor role and mentions manually deleting instances or setting minimum nodes to zero without adequately describing impact and recovery.
The skill identifies major dependencies such as kubectl, jq, and gcloud, documents several failure identifiers, and supplies log and blocker-scanning scripts. Points are deducted because this is a static review with no execution evidence; key paths depend on specific GKE log schemas, the current kube context, tool variants, and permissions, with no committed test suite. Some parsing and fallback paths may fail for complex project, cluster, or resource names.
Trigger terms, target scenarios, and the boundary with the ComputeClass skill are relatively clear, covering scaling, capacity shortages, and troubleshooting. Points are deducted because input/output formats, non-fit scenarios, and the full Autopilot/version/permission boundaries are not explicit. There is no Chinese-language guidance or mainland-China reachability assessment; the core GKE service depends on overseas cloud endpoints whose availability must be confirmed in the user environment.
The structure is readable, with references and assets supporting progressive disclosure. The README documents installation, and repository-level license, official provenance, and issue-reporting paths are clear. Points are deducted because the selected skill lacks its own version, changelog, named maintainer, and update commitment. Installation details, permission matrices, FAQs, and known limitations are incomplete, and several product/version claims require ongoing verification.
The material directly addresses enablement, tuning, scale-up/down blockers, and log diagnosis, with a usable YAML example and operational scripts. Static evidence supports completion of the core guidance task. Points are deducted because no representative outputs or execution results are verified; users must still validate GKE versions, API schemas, IAM, quotas, and actual log fields, and some recommendations may require substantial environment-specific correction.
The skill references repository-local documentation, scripts, and log fields, providing some auditability, and script comments describe expected data shapes. Points are deducted because there is no third-party evidence, committed test suite, CI coverage, or independent reproduction record. External documentation links appear only in the source material, so key version, permission, and product-behavior claims cannot be cross-checked from the supplied files.
- Do not directly apply recommendations involving the Editor role, instance deletion, minimum-node changes, node draining, or cluster updates; confirm targets, permissions, blast radius, and rollback first.
- The scripts read the current kube context and cloud logs and can append to a user-selected path; verify context, project, cluster name, permissions, and log sensitivity before running them.
- Recheck version claims, ComputeClass schemas, the CapacityBuffer CRD, log fields, and IAM recommendations against current official GKE documentation and the target cluster version.
- The supplied material provides no Chinese-language guidance or evidence of reachability from mainland-China networks.
What does this skill do, and when should you use it?
This skill supports Google Kubernetes Engine Cluster Autoscaler, node autoscaling, Node Pool Auto Creation, and Node Auto Provisioning operations. It covers enablement, scale-up and scale-down troubleshooting, consolidation, capacity buffers, zonal stockouts, quotas, and scheduling constraints. It distinguishes modern ComputeClass-based enablement from older cluster-level Node Auto Provisioning. ComputeClass YAML, schemas, and priority configuration are explicitly deferred to the gke-compute-class skill.
Responds to GKE autoscaling questions with gcloud and kubectl guidance; directs users to Cloud Logging entries under container.googleapis.com/cluster-autoscaler-visibility; uses assets/find-scale-down-blockers.sh to scan scale-down blockers; uses assets/log-autoscaler-events.sh <cluster-name> to tail autoscaler events; and points to the CapacityBuffer example for pre-warming serving capacity.
- A GKE operator needs to enable node-pool autoscaling or Node Auto Provisioning.
- A platform engineer is investigating failed scale-ups caused by stockouts, quota, Pod IP exhaustion, or version mismatches.
- A cluster administrator needs to diagnose low-utilization nodes that remain after workloads decline.
- A serving team needs warm standby nodes for traffic spikes without raising min-nodes.
- A stateful-workload owner is analyzing zonal persistent-volume constraints and stockout cooldown cascades.
- A platform team wants to tune consolidation, location policies, Spot handling, or reservation targeting.
What are this skill's strengths and limitations?
- Covers enablement, tuning, scale-up failures, scale-down blockers, capacity buffers, and advanced edge cases.
- Requires a complete scale-down blocker checklist instead of symptom-only troubleshooting.
- Includes log guidance, diagnostic scripts, and a CapacityBuffer YAML example.
- Documents concrete issues involving GKE versions, zonal stockout cascades, reservation cache delay, and topology constraints.
- The source demonstrates guidance and diagnostic assets, not automatic cluster remediation.
- ComputeClass YAML, schemas, and priority configuration are outside its scope.
- The source provides no test suite, platform support matrix, or execution results.
- Several recommendations depend on GKE version, regional capacity, quotas, and cluster configuration.
How do you install this skill?
Run npx skills add google/skills, then select skills/cloud/gke-cluster-autoscaler during installation. The source does not document another installer, a fixed destination directory, or version pinning.
How do you use this skill?
After installation in an Agent Skills-compatible client, ask a concrete operational question such as: “Why are my GKE nodes not scaling down? Check every documented blocker.” For diagnostics, run assets/find-scale-down-blockers.sh [-n namespace] or assets/log-autoscaler-events.sh <cluster-name>. Refer ComputeClass YAML, schemas, and priority settings to gke-compute-class.
How does this skill compare with similar options?
For GKE 1.33.3 and newer, the skill recommends ComputeClasses instead of requiring cluster-level Node Auto Provisioning. Manual node pools use node-pool-level autoscaling. Detailed ComputeClass configuration belongs to gke-compute-class.