Frequently asked questions¶

What is Kubernetes rightsizing?¶

Setting each container's CPU and memory requests and limits to what it actually uses, plus headroom. Requests decide how many pods fit on a node, so over-requested pods waste nodes you pay for, and under-requested pods get throttled or OOM-killed.

Does Kerf change my workloads on its own?¶

Only where you allow it. Every workload starts in audit: Kerf recommends and prices, and changes nothing. You turn on auto-patch per namespace or workload, and CPU and memory can be set separately, so you can start with CPU. See modes. Every change is in the activity log, and a change that makes a workload unhealthy is rolled back (rollback).

VPA already does this. Why Kerf?¶

A Vertical Pod Autoscaler recommends and can apply requests in one cluster. Kerf adds dollars, a fleet UI, per-resource modes, maintenance windows, rollback on bad health and HPA coordination. A VPA is a fine choice for one cluster you watch closely. See the comparison.

Can I use Kerf with the Horizontal Pod Autoscaler?¶

Yes. When an HPA scales a workload on CPU or memory utilization, Kerf changes the requests and the HPA's utilization target in one change, so the HPA keeps scaling at the same usage as before. See HPA coordination.

Does Kerf send data anywhere?¶

No usage data leaves your infrastructure. Kerf reads metrics from your Prometheus-compatible store and keeps its state in your PostgreSQL. Two kinds of calls go outside your clusters:

the AWS Pricing and EC2 APIs, only when you choose AWS prices;
a daily check of the public GitHub releases of kerf-dev/kerf for a newer version. It sends nothing about your installation. Turn it off with hub.updateCheck.enabled: false (maintenance).

Does it work outside EKS?¶

The recommender, agent and webhook use standard Kubernetes APIs. EKS 1.35 is the tested platform. On other clusters, prices come from a static table instead of the AWS price list. See install on any cluster.

What does it need to run?¶

One Helm chart. The hub needs PostgreSQL and a Prometheus-compatible metrics endpoint (Prometheus, VictoriaMetrics, Thanos, Mimir or Amazon Managed Service for Prometheus) that you probably have already. Workload clusters need cert-manager for the webhook. See Install on Amazon EKS.

Does it work with Argo CD?¶

Yes. An in-place resize does not touch the pod template, so Argo CD sees no drift. Policies can live in Git as RightsizingPolicy resources. See GitOps and Argo CD.

A recommendation or a change did not happen. Why?¶

The workload's mode, a guardrail, a maintenance or freeze window, or the cluster's readiness may be holding it back. Start with troubleshooting and the readiness checks.

Is it ready for production?¶

Kerf is new (0.x). Run it in audit first, then turn on auto-patch for one namespace and watch the activity log. Until 1.0.0, a minor version may change the chart values or the custom resources; the changelog says how to upgrade.

What does it cost?¶

Nothing. Kerf is open source under the Apache License 2.0, with no paid tier and no hosted service. You pay for the compute it runs on.