Right-Sizing Your Kubernetes Workloads: The Ultimate

In Kubernetes, "over-provisioning" is a silent budget killer. Assigning 1 CPU to a microservice that only uses 25m is like paying for a skyscraper when you only need a studio apartment. Goldilocks is the perfect tool to find your "just right" resource limits.

Here is your 2026 step-by-step guide to a perfectly sized cluster.

Step 1: Install the Metrics Server (The Eyes)

VPA cannot make recommendations if it doesn't know how much CPU/RAM your pods are currently using.

Standard Installation:

⚠️ Critical Fix for Local Labs (KIND, Minikube, etc.)

In local environments, you will likely see a tls: failed to verify certificate error because Metrics Server cannot validate the Node's IP. To fix this, you must enable the insecure TLS mode.

Run this command to patch the deployment immediately:

Alternatively, use kubectl edit deployment metrics-server -n kube-system and add --kubelet-insecure-tls under the args section.

Verification: Run kubectl top nodes. If you see CPU/Memory percentages, your "eyes" are working!

Step 2: Install Vertical Pod Autoscaler (The Brain)

Goldilocks visualizes data provided by the Vertical Pod Autoscaler (VPA). You need the VPA components to analyze usage patterns.

Install via Helm:

Step 3: Install Goldilocks (The Dashboard)

Now, we install the "face" of the operation to turn complex VPA data into easy-to-read charts.

Install via Helm:

Step 4: Activating Analysis

Goldilocks follows an opt-in model. It will only analyze namespaces you specifically label. To start the magic for your game or production namespace:

Accessing the Results

Since you are likely on a local machine, use Lens to port-forward or run:

Visit http://localhost:8080 to see your recommendations.