RB018: Container registry cleanup¶
Applicability: GCP Artifact Registry primary; AWS ECR optional. Operator: release/platform owner. Verified: 2026-10-23.
Symptom¶
Registry storage cost or image count grows, cleanup policy fails, or old images appear eligible for deletion.
Investigation¶
- Inventory repositories, digests, tags, creation/download time, size, and retention policy. Work only from digests; tags are mutable references.
- Build a protected set from running portal/worker/provisioner/Guacamole workloads, current and rollback releases, release evidence, provisioner inputs, and image-promotion channels.
- On AWS, Terraform lifecycle policy retains the configured recent image set. Confirm policy status and exclusions before manual action.
- GCP Artifact Registry currently has no repository-owned cleanup policy. A list of old images is evidence, not authorization to delete them.
Resolution¶
For AWS, repair and apply the Terraform-owned lifecycle policy. For GCP, preserve the inventory and escalate a reviewed cleanup-policy change; do not perform ad hoc deletion from this runbook. Any approved emergency deletion must use exact digests, exclude the protected set, be bounded and logged, and verify that running/rollback images still resolve.
Prevention¶
Add retention as provider-owned Terraform with dry-run reporting, explicit protected release classes, cost thresholds, and rollback tests. Track registry growth and unreferenced digests before the budget limit is reached.
Escalate every GCP deletion until a qualified automated policy exists, or when provenance cannot establish whether a digest is still required.