What this fixes

A BackendTrafficPolicy that resolves to zero targets currently keeps a stale status. This addresses two issues with the same symptom but different root causes:

• Fixes BackendTrafficPolicy status does not report missing targetRef and leaves stale observedGeneration #8926 — targetRef/targetRefs pointing at a nonexistent object. The policy stays in the result set carrying its deep-copied prior status, so Accepted=True persists with an out-of-date observedGeneration.
• Fixes BackendTrafficPolicy status becomes stale when targetSelectors change from matching routes to matching no routes #8927 — targetSelectors that match nothing. The policy never enters the result set, so its status is never reconciled and stays stale.

The approach

Per review feedback on #9416, the stale status is cleaned up in the provider status updater rather than by synthesizing a condition in the translator:

• Translator (ProcessBackendTrafficPolicies): stops propagating carried-over stale status for a policy that attached to zero targets this pass. This is subtractive — no synthesized condition and no invented ancestorRef. Both cases then produce no status for the policy, which the gatewayapi runner turns into a status-store delete.
• Provider (internal/provider/kubernetes/status.go, BackendTrafficPolicy updater): handles the resulting status delete by stripping the ancestor conditions owned by this controller from the live object, while preserving ancestors owned by other controllers. If the object itself was deleted, the status updater's Get returns NotFound and applying the update is a no-op.

Scope & follow-up

This PR is intentionally scoped to BackendTrafficPolicy to stay focused on the two filed issues. The same two structural gaps exist identically across the sibling policy types (SecurityPolicy, ClientTrafficPolicy, BackendTLSPolicy, EnvoyExtensionPolicy, ExtensionServerPolicy):

1. the translator's early-return-without-status on a missing/zero target, and
2. the if update.Delete { return } skip in each status updater (the route path even carries a TODO about the same dangling-status problem).

Both halves here were written to be extractable into shared helpers, so if the maintainers are happy with the approach and cleanup semantics, I'm glad to generalize it across the other policy types in a follow-up PR.

Testing

• New golden fixtures reproducing each case:
  • backendtrafficpolicy-targetref-not-found (BackendTrafficPolicy status does not report missing targetRef and leaves stale observedGeneration #8926)
  • backendtrafficpolicy-targetref-not-found-stale-status (BackendTrafficPolicy status does not report missing targetRef and leaves stale observedGeneration #8926 with a pre-existing on-object stale status)
  • backendtrafficpolicy-selector-no-match (BackendTrafficPolicy status becomes stale when targetSelectors change from matching routes to matching no routes #8927)
• A new end-to-end provider test drives the delete-path cleanup through the real UpdateHandler.apply against a fake client: stale status owned by our controller is cleared, another controller's status is preserved, and a delete for an already-removed object is a no-op.
• go test ./internal/gatewayapi/... ./internal/provider/kubernetes/... passes; go vet and gofmt are clean. Regenerating goldens changed no existing files.

Release note

Added release-notes/current/bug_fixes/8926-btp-status-target-not-found.md.

Made with Cursor

Made with Cursor