Post

Enrolling a Native Argo CD Image Updater Canary

Enrolling a Native Argo CD Image Updater Canary

I deployed a native Argo CD Image Updater canary to validate the controller’s write-back flow before enrolling real workloads. The canary tracks the Harbor mirror’s curl 8.22.0 tag using the native digest strategy, starts from a reviewed 8.21.0 digest, and runs a sleeping container in the automation namespace with all traffic denied.

How the canary exercises the native write-back path

The upstream Helm chart runs a single Image Updater controller replica in the argocd namespace that watches explicit ImageUpdater resources. The canary ApplicationSet syncs one ImageUpdater and one generated Application. The ImageUpdater points at the Harbor mirror’s curl 8.22.0 tag with the native digest update strategy. The generated Application’s Deployment starts from a reviewed curl 8.21.0 digest. The ApplicationSet is configured to ignore the generated Application’s Kustomize image override so the controller can write the resolved digest back through Argo CD. The workload itself is a single sleeping container in the automation namespace with both inbound and outbound traffic denied, so the canary tests the update mechanics without exposing a network surface.

Verification commands for the controller result

The documentation now includes three kubectl commands to confirm the loop works: check the ImageUpdater status in argocd, inspect the generated Application’s Kustomize image override, and read the live pod’s ready state, image reference, and image ID in the automation namespace. These commands let me verify that the controller resolved the tag, wrote the digest to the Application, and that Argo CD rolled the pod to the new image.

Where this canary pattern transfers

A platform team can use the same approach when adopting any GitOps image automation: deploy a harmless workload that exercises the full controller-to-Application-to-pod path, lock its network access, and define the exact verification steps. Once the canary demonstrates reliable digest resolution and write-back, the same ImageUpdater resource pattern can be applied to production workloads with confidence that the control loop behaves as documented.

This post is licensed under CC BY 4.0 by the author.