fix: disambiguate OLM Subscription from ACM Subscription in oc calls

[coreywan]

Problem

tools/cluster_setup/openshift_with_stack.sh calls oc get subscription / oc delete subscription using the short resource name, which assumes it resolves to subscriptions.operators.coreos.com.

On a cluster that also runs Red Hat Advanced Cluster Management, ACM registers a second CRD under the same plural name, and oc resolves the ambiguity to the ACM group:

$ oc api-resources | grep '^subscriptions'
subscriptions   appsub     apps.open-cluster-management.io/v1   true   Subscription
subscriptions   sub,subs   operators.coreos.com/v1alpha1        true   Subscription

$ oc get subscription nfd -n openshift-nfd
Error from server (NotFound): subscriptions.apps.open-cluster-management.io "nfd" not found

Impact

1. Install hangs, then fails with a misleading error. wait_for_subscription_csv() discards the NotFound via 2>/dev/null || true, so current_csv and installed_csv come back empty on every poll. csv_name stays empty, the CSV phase check is never reached, and the loop spins for the full timeout:

[INFO] Waiting for OLM Subscription openshift-nfd/nfd (timeout: 600s)...
<600s of nothing>
[ERROR] Timed out waiting for OLM Subscription openshift-nfd/nfd
        (currentCSV=not-created, installedCSV=not-installed, target phase=unknown)

The operator had in fact installed correctly the whole time:

$ oc get subscriptions.operators.coreos.com nfd -n openshift-nfd     -o jsonpath='{.status.installedCSV}{"\n"}{.status.state}'
nfd.4.21.0-202608241535
AtLatestKnown

$ oc get csv -n openshift-nfd
nfd.4.21.0-202608241535   Node Feature Discovery Operator   Succeeded

$ oc get pods -n openshift-nfd
nfd-controller-manager-689844599c-vpbvs   1/1   Running   0   3h15m

This blocks step 2 (NFD, 600s) and would block step 3 again (GPU Operator, 900s).

2. Teardown leaves orphans. main_delete() pairs the same lookup with --ignore-not-found=true, so the OLM Subscription and its CSV are silently left behind on delete.

Fix

Pin all nine call sites to the fully qualified subscriptions.operators.coreos.com. No logic changes.

This is a no-op on clusters without ACM, where the short name already resolves to the OLM group. openshift_with_stack.sh is the only file in the repo using the ambiguous short name.

Testing

• Reproduced on OpenShift 4.21.30 with ACM present: install blocked at step 2 with NFD fully healthy.
• With the patch applied, wait_for_subscription_csv returns on its first poll and the install proceeds to step 3.
• bash -n tools/cluster_setup/openshift_with_stack.sh passes.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1b78cd500f

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Update the OLM readiness mock for the qualified resource

When tools/cluster_setup/test_openshift_with_stack.sh runs, its oc mock at lines 603–606 only returns CSV names when ${2:-} equals subscription. These qualified lookups therefore return nothing, causing wait_for_subscription_csv to exhaust all three mocked polls; the assertion at lines 616–617 receives 3 instead of 1, and the test suite exits nonzero. Update the mock to recognize subscriptions.operators.coreos.com alongside this production change.

Useful? React with 👍 / 👎.