Pre-requisites¶

Install these before Installation. Missing any item usually shows up as empty recommendations, failing health checks, or webhook errors.

Cluster and tooling¶

Requirement	Notes
Kubernetes 1.33+	In-place pod resource updates are part of the design; older versions are unsupported. PSI-aware optimization requires 1.34+; see Prometheus section below.

kubectl	Configured for the target cluster context.
Helm 3	For installing the official chart (OCI registry).

Prometheus¶

CruiseKube reads container and node metrics (usage, throttling, PSI where exposed, etc.) from Prometheus.

You may use an existing Prometheus or install one via Helm (given below).
Set CRUISEKUBE_DEPENDENCIES_INCLUSTER_PROMETHEUSURL (or equivalent) to a URL reachable from the controller pods (in-cluster Service URL, not localhost).

Optional: install Prometheus via Helm

If you do not already have Prometheus, you can add kube-prometheus-stack. A second install can fail if Prometheus already exists in the namespace—reuse the existing instance instead.

helm repo add prometheus-community https://prometheus-community.github.io/helm-charts
helm repo update
helm install kube-prometheus-stack prometheus-community/kube-prometheus-stack \
  --namespace monitoring \
  --create-namespace \
  --set alertmanager.enabled=false \
  --set grafana.enabled=false \
  --set prometheus.prometheusSpec.serviceMonitorSelectorNilUsesHelmValues=false

PSI (Pressure Stall Indicator): CruiseKube’s algorithm is built around PSI-aware CPU reasoning on clusters that expose the right metrics (Kubernetes 1.34+ PSI story). If PSI is absent, behavior degrades toward usage-only signals—still useful, but not identical to a full PSI deployment. See Algorithm.

PostgreSQL¶

CruiseKube persists workload statistics, recommendations, and per-workload overrides in a database.

Option A: Bitnami PostgreSQL subchart official Helm chart (postgresql.enabled=true), is enabled by default.
Option B: Use your own Postgres and set global.postgresql.auth.* (host, port, user, password, database) per Helm chart reference.

Network and RBAC¶

Controller and webhook must reach kube-apiserver, Prometheus, and PostgreSQL.
The chart installs RBAC and MutatingWebhookConfiguration resources; ensure your GitOps / policy engines allow them.

What you do not need (for a minimal install)¶

Grafana (optional for you; not required by CruiseKube).
A separate metrics long-term store (CruiseKube queries Prometheus directly).