GKE Workload Autoscaling
Configure manual scaling, HPA, and VPA for GKE workloads while improving resource requests from observed usage.
The skill clearly scopes itself to GKE workload autoscaling and discloses restart and eviction risks for HPA/VPA; however, kubectl scale, autoscale, gcloud update, and VPA Auto operations mutate an external cluster without explicit confirmation, least-privilege guidance, or rollback procedures, so points are deducted. No credential handling or covert exfiltration is shown.
The workflows, manifests, prerequisites, and verification commands are broadly consistent and cover manual scaling, HPA, VPA, and rightsizing; however, there is no abnormal-input handling, diagnostic failure feedback, compatibility validation, or test evidence. Several GKE version and Metrics Server claims cannot be confirmed from static files, limiting the score.
The name, audience, scenarios, and exclusions for Cluster Autoscaler, static sizing, and direct node-machine configuration are clear; however, required inputs, environment assumptions, non-fit boundaries, and semantic trigger details remain incomplete. Chinese-language support and mainland-China reachability are not addressed, so points are deducted.
The documentation is readable and layered through workflows, best practices, and asset templates. The repository supplies installation guidance, an Apache-2.0 license, issue-reporting route, and an active-development note. The skill lacks its own version, changelog, explicit maintainer, and thorough troubleshooting; the README link also differs from the assessed path name, so points are deducted.
The commands and HPA/VPA YAML templates provide a useful starting point for the core scaling tasks, with a rightsizing workflow included; however, the templates use generic names and the default namespace, and production checks, rollback, and verification coverage are incomplete. Static review cannot establish directly runnable results, so the score follows the static ceiling.
The supplied skill files, example manifests, README, license, and pinned revision are auditable evidence; however, there is no committed key-path test suite, CI execution evidence, or independent reproduction. The conclusion therefore rests mainly on static documentation and receives limited credit.
- Before applying scaling changes or enabling VPA, confirm the target cluster, namespace, deployment, and side effects, and prepare a verifiable rollback plan.
- Validate the GKE version, Metrics Server, VPA support, and HPA/VPA metric interactions; the generic names and default namespace in the examples are not production-ready by themselves.
- The README links to gke-scaling while the assessed path is gke-workload-scaling; verify the path mapping during discovery or installation.
What does this skill do, and when should you use it?
This skill focuses on scaling application workloads running on Google Kubernetes Engine. It covers fixed replica changes for Deployments, Horizontal Pod Autoscaling (HPA), and Vertical Pod Autoscaling (VPA) for resource right-sizing. It includes kubectl and gcloud examples, YAML manifest references, prerequisites, update modes, and operational best practices. It explicitly excludes Cluster Autoscaler, static cluster sizing, and direct node machine-style configuration.
Guides users through kubectl scale for fixed Deployment replicas, kubectl autoscale or an HPA manifest for CPU, memory, or custom-metric scaling, and kubectl get commands for verification. It shows how to enable VPA on a Standard GKE cluster with gcloud and explains Off, Initial, Auto, and InPlaceOrRecreate update modes. It also describes reading VPA recommendations with kubectl describe vpa, comparing target values with current requests, and applying a 20% buffer when calculating new requests.
- A GKE operator needs to set a Deployment to a fixed replica count for immediate intervention or testing.
- A platform engineer wants Pods to scale horizontally from CPU, memory, or custom metrics.
- A Kubernetes team needs to right-size CPU and memory requests based on observed workload usage.
- A team wants to run VPA in Off mode for at least 24 hours before reviewing recommendations and changing manifests.
What are this skill's strengths and limitations?
- Covers manual scaling, HPA, and VPA in one focused GKE workflow.
- Provides copyable kubectl and gcloud commands plus referenced HPA and VPA YAML templates.
- Addresses resource requests, metric conflicts, PDBs, HPA stabilization, and VPA restart risks.
- Includes a concrete P95-based right-sizing process with a 20% buffer.
- Does not cover Cluster Autoscaler, static cluster capacity, or node machine configuration.
- HPA requires Metrics Server, and reliable HPA/VPA behavior depends on accurate resource requests.
- VPA Auto mode can restart running Pods, so applications must handle termination gracefully.
- The supplied material does not include a test suite, compatibility matrix, or complete cluster authentication procedure.
How do you install this skill?
Run npx skills add google/skills and select the desired skill during installation. The README does not document the exact installation directory or command-line selection flags; the target file is skills/cloud/gke-workload-scaling/SKILL.md.
How do you use this skill?
In an Agent Skills-compatible client, use a request such as “Configure HPA for the GKE deployment deployment_name with 1–10 replicas and check its resource requests.” The workflow assumes access to a GKE cluster and kubectl; VPA enablement on Standard clusters also uses gcloud. Authentication steps are not documented in the supplied source.
How does this skill compare with similar options?
The skill positions HPA as the mechanism for changing Pod count and VPA as the mechanism for changing Pod CPU and memory resources. It recommends avoiding the same metric for both. Cluster Autoscaler is named as a separate area that this skill does not address.