Skip Route lookups during cleanup on clusters without the Route API

[lwr20]

Problem

On a plain-Kubernetes cluster the ForkliftController never reaches Successful. It settles at:

Successful=False   Failure=True (Failed)   Running=False
message: MODULE FAILURE: No start of json char found   (x3)
ansibleResult: {"changed":0,"failures":0,"ok":62,"skipped":62}

The install itself is fine — 6/6 deployments Available, 6/6 cert-manager Certificates and both Issuers Ready=True. Only the status is wrong.

Cause

forklift_resources (defaults/main.yml:102) is [Deployment, ConfigMap, Service, Route], and the four disabled-feature cleanup loops in tasks/main.yml iterate it unguarded. The enable paths are all correctly gated on not k8s_cluster|bool (lines 325/370/393); the cleanup loops are not.

So the documented plain-Kubernetes CR — which sets feature_ui_plugin: "false", feature_cli_download: "false" — drives cleanup for those features, and cleanup.yml calls k8s_info with kind: Route. With no route.openshift.io on the cluster, the dynamic client recurses:

File ".../kubernetes/dynamic/resource.py", in __search
    return self.__search(parts[1:], resourcePart, reqParams + [part])
RecursionError: maximum recursion depth exceeded

The module writes a traceback to stderr and nothing to stdout, so Ansible reports the opaque MODULE FAILURE: No start of json char found, naming neither Route nor the recursion.

The precise trigger is an unqualified kind — cleanup.yml passes kind with no api_version, so the dynamic client searches every group and __search recurses without terminating when nothing matches. Verified in the operator image against this cluster:

k8s_info kind=Route                                              -> FAILED, RecursionError
k8s_info api_version=console.openshift.io/v1 kind=ConsolePlugin  -> SUCCESS, "Failed to find API for resource..."

A kind qualified with its api_version fails gracefully. This is worth knowing for two reasons: the finalizer's Remove console plugin is not affected (its template names apiVersion: console.openshift.io/v1), and the fix below addresses the current resource list rather than the underlying class — any future OpenShift-only kind added to forklift_resources would reintroduce it. Carrying api_version alongside each kind would be the structural fix, but the four kinds span different groups, so that is a larger change than this release needs.

cleanup.yml already has a rescue: for "empty or missing resources" and it works — hence failures: 0 and a healthy install. What it cannot suppress is the operator recording the module-failure events, which is what pins the CR at Failure=True.

Confirmed deterministic: three failing tasks per reconcile (Get Route resources labeled forklift-{ui-plugin,cli-download,mcp-server}), 13 distinct Ansible job IDs over ~3h, identical every time.

Impact

• The documented install step kubectl wait --for=condition=Successful forkliftcontroller/... --timeout=300s always times out.
• Any automation gating on Successful hangs.
• Failure/Successful stop being usable health signals on Kubernetes — a genuine later failure is indistinguishable from this noise.

Fix

Filter Route out of the cleanup kinds when the API group is absent, reusing the api_groups detection the role already does two lines above. Keying off api_groups rather than k8s_cluster means the decision follows what the cluster can actually resolve, not a user-settable flag.

Testing

RKE2 v1.33.5, 5 nodes, Calico Enterprise v3.24.0-2.0-calient-1.dev-151, Forklift v2.12.5-v3.24.0-2.0, installed from the published non-OLM manifests.

• Reproduced before the change, as above.
• Jinja verified both ways: with route.openshift.io present the list is unchanged; absent, it is [Deployment, ConfigMap, Service].
• A full vSphere -> Kubernetes migration succeeds on this cluster either way (the install is functional; only the CR status is affected).

Note

cleanup.yml and the forklift_resources list are byte-identical to kubev2v/forklift@main, so this affects every non-OpenShift Forklift user and is worth raising upstream too. This PR fixes it on the release branch.

[Copilot]

Pull request overview

This pull request fixes a reconcile-time failure on plain Kubernetes clusters by preventing the disabled-feature cleanup tasks from querying OpenShift-only Route resources when the route.openshift.io API group is not available, avoiding Ansible k8s_info module failures that incorrectly pin the CR status to Failure=True.

Changes:

• Introduces a derived cleanup_resources list based on detected cluster API groups, filtering out Route when route.openshift.io is absent.
• Updates all disabled-feature cleanup loops to iterate cleanup_resources instead of forklift_resources.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.