Skip to main content

Installation Overview

K8s Sentinel offers three installation methods. Choose the one that best fits your use case.

Comparison​

MethodUse CaseRequirementsAdvantages
Kubernetes JobProduction / CI pipelineskubectl accessAutomated, no manual steps
Local Script (install.sh)Quick evaluation / developmentbash + kubectlVisible steps, easy to observe
Offline Installation (v0.71+)The cluster has no internet accessA private registry + kubectlNo Helm and no outbound access needed on the cluster side

Common Prerequisites​

Regardless of installation method, confirm the following before starting:

  • Kubernetes cluster is running (version 1.32+)
  • kubectl is installed and configured with cluster-admin permissions
  • Cilium is deployed with kube-proxy replacement and Hubble enabled (see Installing and Configuring Cilium)
  • Tetragon is installed (see Installing Tetragon)

Complete all checks in the Prerequisites page before proceeding.

Re-applying deploy/sentinel.yaml puts back what the file says

Most upgrades require re-applying deploy/sentinel.yaml. If you changed the Deployment or the Service by editing that file (a PVC in place of the emptyDir, AUDIT_WEBHOOK_TOKEN, a pinned image tag, a NodePort), carry those edits into the new file first, or apply your own copy.

Applied as it comes, the data volume goes back to emptyDir, which is an empty K8s Sentinel with only the admin / admin account, and an audit webhook token set in the file is dropped. Changes made with kubectl set env or kubectl patch, and a clusterIP you chose, are kept.

Upgrading to v0.66 through v0.70 (refusing vulnerable images)
  • Re-apply deploy/sentinel.yaml, then restart K8s Sentinel: it adds the VulnerabilityGate kind, which v0.67, v0.68 and v0.70 each extended. Without it the matching admission policies admit everything; see Refusing Vulnerable Images for what each version shows
  • The Deployment Gate page of v0.65.0 is gone, and the policy it made is removed on upgrade; make it again under Admission Policy
  • Vulnerability Limit, Blocked CVEs, Host Path Volumes and Environment Variables policies made before v0.69 are best made again in the form, see Admission Policy
  • v0.71 and v0.72 have no manifest change
Upgrading from before v0.43?

Since v0.43, runtime events are collected over gRPC (no more kubectl exec). Existing clusters need three things before upgrading:

  1. Expose Tetragon's gRPC: kubectl -n kube-system patch cm tetragon-config --type merge -p '{"data":{"server-address":"0.0.0.0:54321"}}' then rollout restart ds/tetragon (see Installing Tetragon)
  2. Enable Hubble Relay on Cilium: --set hubble.relay.enabled=true (see Installing and Configuring Cilium)
  3. Re-apply the ClusterRole: kubectl apply -f deploy/sentinel.yaml - the new version drops pods/exec (the largest privilege K8s Sentinel held) and adds read access to the resources exposure detection needs

Fresh installs (install-job / install.sh) handle all of this automatically.

Upgrading to v0.64+
  • Everyone is signed out once: since v0.64 a session is refused after its account is deleted or its password changes
  • Creating a Tracing Policy under a name that already exists is refused instead of replacing the existing policy
  • Re-apply deploy/sentinel.yaml to drop the unused cluster-wide pods/log permission (optional)
  • An existing Tetragon is still on the host network: since v0.64.1 the installers put Tetragon on the pod network and limit its gRPC port to K8s Sentinel with a NetworkPolicy, but they do not change a Tetragon that is already installed. See Installing Tetragon for how to move it
Upgrading to v0.60+ (vulnerability scanning)
  • Re-apply deploy/sentinel.yaml: the ClusterRole adds read-only rules for vulnerabilityreports and jobs (v0.60), plus a Role in sentinel-system used by registry scanning (v0.62)
  • Trivy Operator: it is not installed automatically on existing environments; see Installing Trivy Operator for how to install and upgrade it

After installation:

Recommendation​

Which method should I use?
  • Production environments: Use Kubernetes Job Installation. The Job runs inside the cluster, ensuring network consistency and making it easy to audit and integrate with automation.
  • Quick evaluation or development: Use Local Script Installation. The script prints installation progress step by step, making it easy to observe and debug.
  • A cluster with no internet access: Use Offline Installation. Carry the images to a private registry first, then install from the offline bundle.