Skip to content

Deploying and testing AppControl on OpenShift

The Helm chart in helm/appcontrol is the single source of truth for both Kubernetes and OpenShift. The OpenShift profile adds a TLS Route, disables Ingress, and leaves UID/GID selection to the namespace Security Context Constraint (SCC).

What is tested

The OpenShift GitHub Actions workflow has two levels:

  1. Every relevant pull request renders and lints the chart, checks the Route, and rejects fixed UIDs/GIDs, privileged containers, or incomplete restricted-v2 security contexts.
  2. Manual and scheduled runs deploy AppControl to a real OpenShift cluster, verify the SCC selected by admission, deploy an agent, and execute the same start/check/stop/restart scenario as the local full-stack E2E test.

The CI does not install an OpenShift cluster inside a GitHub-hosted runner. OpenShift Local requires a virtual machine and is not suitable for nested virtualization there. Instead, the workflow connects to an existing cluster.

Red Hat Developer Sandbox

The no-cost Red Hat Developer Sandbox is suitable for this E2E job. It is a shared, quota-limited trial environment, so use one dedicated project and let the workflow remove its release after every run.

  1. Start a sandbox from Red Hat Developer Sandbox.
  2. In the OpenShift console, open Copy login command, then Display Token.
  3. Record the three values shown by:
oc whoami --show-server
oc whoami -t
oc project -q
  1. In the GitHub repository, create these Actions secrets:
Secret Value
OPENSHIFT_SERVER Output of oc whoami --show-server
OPENSHIFT_TOKEN Output of oc whoami -t
OPENSHIFT_NAMESPACE Output of oc project -q

Tokens and sandbox instances expire. Replace OPENSHIFT_TOKEN when login starts failing, and update all three values when a new sandbox is provisioned.

Launch the CI deployment

After the workflow is present on GitHub:

  1. Open Actions → OpenShift.
  2. Select Run workflow.
  3. Select the branch to test.
  4. Follow the Real OpenShift deployment and execution job.

The normal CI / Docker Build matrix builds backend, frontend, gateway, and agent once, in parallel, and publishes immutable SHA-tagged images to GHCR. The OpenShift job verifies and consumes those exact images; it does not rebuild them. It then creates short-lived secrets, installs the chart, runs the E2E scenario, uploads diagnostics on failure, and removes only resources belonging to its appcontrol-e2e release.

Scheduled workflows run only from the default branch. The weekly schedule therefore becomes active after this change is merged into main.

Manual Helm deployment

Production deployments should use an external or operator-managed PostgreSQL database. Create secrets without placing credentials in a values file:

oc create secret generic appcontrol-database \
  --from-literal=database-url='postgresql://USER:PASSWORD@HOST:5432/appcontrol'

oc create secret generic appcontrol-jwt \
  --from-literal=jwt-secret="$(openssl rand -hex 48)"

Deploy:

helm upgrade --install appcontrol helm/appcontrol \
  --namespace "$(oc project -q)" \
  -f helm/appcontrol/values-openshift.yaml \
  --set backend.database.existingSecret=appcontrol-database \
  --set backend.jwt.existingSecret=appcontrol-jwt \
  --set backend.image.repository=REGISTRY/appcontrol-backend \
  --set frontend.image.repository=REGISTRY/appcontrol-frontend \
  --set gateway.image.repository=REGISTRY/appcontrol-gateway \
  --set backend.image.tag=VERSION \
  --set frontend.image.tag=VERSION \
  --set gateway.image.tag=VERSION \
  --wait --timeout 12m

Retrieve the generated Route:

oc get route appcontrol -o jsonpath='https://{.spec.host}{"\n"}'

Security model

The chart is designed for restricted-v2 and restricted-v3:

  • no fixed runAsUser, runAsGroup, or fsGroup;
  • runAsNonRoot: true;
  • seccompProfile: RuntimeDefault;
  • privilege escalation disabled;
  • all Linux capabilities dropped;
  • application images permit an arbitrary OpenShift UID to access only the files and writable directories it needs.

The frontend proxies /api, /ws, /health, and /ready internally. Only one edge-terminated Route is required for the browser application.

Diagnostics

oc get pods,route,pvc
oc get events --sort-by=.lastTimestamp
oc logs deployment/appcontrol-backend --tail=200
oc logs deployment/appcontrol-gateway --tail=200
oc describe pod POD_NAME

For CI failures, download the openshift-diagnostics artifact from the failed GitHub Actions run.