test: report tier reference proof setup failures without hanging

.config/e2e-full-selection.txt

@@ -1,2 +1,2 @@
 sha256-darwin=874c881d7b45f12378a5817c7f42c95c4981960a2ec9ce12dcf4af239ae1f9d5
-sha256-linux=9351e25b45bf7dfce18b951a5e3740225f457cacc53b8bf9f500f6947763ec0e
+sha256-linux=9515861be899ceb10e2e0ef93c34208bb7a7a8a7f8067a02db4cfba23270ebd6

crates/ecstore/src/services/tier/tier.rs

@@ -17540,6 +17540,66 @@ mod tests {
         assert!(current.tiers.contains_key("COLD-B"));
     }
 
+    async fn wait_for_reference_proof_barrier(
+        barrier: &TierDriverBuildBarrier,
+        update: &mut tokio::task::JoinHandle<std::result::Result<(), TierConfigUpdateError>>,
+    ) -> std::result::Result<(), String> {
+        tokio::select! {
+            biased;
+            result = &mut *update => Err(format!("tier update exited before the reference proof barrier: {result:?}")),
+            () = barrier.arrived.notified() => Ok(()),
+            () = tokio::time::sleep(Duration::from_secs(30)) => {
+                // Aborting the caller does not stop its owned mutation task.
+                // Let a late arrival pass the test-only barrier.
+                barrier.release.add_permits(1);
+                update.abort();
+                Err("timed out waiting for the reference proof barrier".to_string())
+            }
+        }
+    }
+
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn reference_proof_barrier_reports_update_failure_before_arrival() {
+        let manager = TierConfigMgr::new();
+        let store = Arc::new(CasConfigStore::default());
+        let mut persisted = empty_mgr();
+        persisted.tiers.insert("COLD-A".to_string(), build_rustfs_tier("COLD-A"));
+        persisted
+            .save_tiering_config_if_current(store.clone(), None)
+            .await
+            .expect("early update failure fixture should persist");
+        let barrier = tier_reference_proof_test_barrier();
+        let scoped_barrier = barrier.clone();
+        let factory: TierDriverTestFactory =
+            Arc::new(|_| Err(AdminError::msg("injected driver initialization failure before reference proof")));
+        let mut update = tokio::spawn(async move {
+            TIER_REFERENCE_PROOF_TEST_BARRIER
+                .scope(
+                    scoped_barrier,
+                    TIER_DRIVER_TEST_FACTORY.scope(
+                        factory,
+                        TIER_MUTATION_TEST_PEERS.scope(
+                            Vec::new(),
+                            TierConfigMgr::update_candidate_with_config_lock(
+                                &manager,
+                                store,
+                                TierCandidateMutation::Remove("COLD-A".to_string(), true),
+                            ),
+                        ),
+                    ),
+                )
+                .await
+        });
+
+        let err = tokio::time::timeout(Duration::from_secs(5), wait_for_reference_proof_barrier(&barrier, &mut update))
+            .await
+            .expect("an early update failure should be observed without waiting for the barrier deadline")
+            .expect_err("a failed update cannot reach the reference proof barrier");
+        assert!(err.contains("Mutation"), "{err}");
+        assert!(err.contains("injected driver initialization failure before reference proof"), "{err}");
+    }
+
     #[tokio::test]
     #[serial_test::serial]
     async fn reference_proof_rejects_a_changed_prepared_fence_revision_before_publish() {
@@ -17560,7 +17620,7 @@ mod tests {
         let scoped_barrier = barrier.clone();
         let update_manager = manager.clone();
         let update_store = store.clone();
-        let update = tokio::spawn(async move {
+        let mut update = tokio::spawn(async move {
             TIER_REFERENCE_PROOF_TEST_BARRIER
                 .scope(
                     scoped_barrier,
@@ -17575,7 +17635,9 @@ mod tests {
                 )
                 .await
         });
-        barrier.arrived.notified().await;
+        wait_for_reference_proof_barrier(&barrier, &mut update)
+            .await
+            .expect("tier update should reach the reference proof barrier");
 
         let unrelated = prepared_remove_intent("COLD-B", uuid::Uuid::from_u128(0x2237));
         TierConfigMgr::apply_prepared_mutation_intent_block(&manager, &unrelated)
@@ -17583,8 +17645,9 @@ mod tests {
             .expect("an unrelated prepared fence should advance the runtime revision");
         barrier.release.add_permits(1);
 
-        let err = update
+        let err = tokio::time::timeout(Duration::from_secs(30), update)
             .await
+            .expect("tier update should finish after the reference proof barrier releases")
             .expect("tier update task should join")
             .expect_err("a reference proof cannot authorize publication across a fence revision change");
         let TierConfigUpdateError::Publish(err) = err else {

scripts/ci_gate.py

@@ -59,7 +59,7 @@ def expected_results(mode: str, event: str, ref: str) -> dict[str, str]:
     expected.update({job: "success" if mode == "full" else "skipped" for job in CODE_JOBS})
     rio = mode == "full" and event in ("schedule", "workflow_dispatch")
     expected.update({job: "success" if rio else "skipped" for job in OPTIONAL_JOBS[:2]})
-    full = mode == "full" and (event in ("merge_group", "workflow_dispatch") or (event == "push" and ref in ("refs/heads/main", "refs/heads/release")))
+    full = mode == "full" and (event in ("merge_group", "workflow_dispatch") or (event == "push" and ref == "refs/heads/main"))
     expected["e2e-full"] = "success" if full else "skipped"
     return expected
 
@@ -219,27 +219,6 @@ class SelfTests(unittest.TestCase):
                 bad = {**good, "classify-changes": {"result": "success", "outputs": selection}}
                 self.assertTrue(verify_results(bad, event, "refs/heads/main"))
 
-    def test_full_e2e_gate_preserves_workflow_branch_and_event_scope(self):
-        for event, ref, required in (
-            ("push", "refs/heads/main", "success"),
-            ("push", "refs/heads/release", "success"),
-            ("push", "refs/heads/feature", "skipped"),
-            ("push", "refs/heads/release-candidate", "skipped"),
-            ("push", "refs/tags/release", "skipped"),
-            ("pull_request", "refs/pull/1/merge", "skipped"),
-            ("schedule", "refs/heads/release", "skipped"),
-            ("workflow_dispatch", "refs/heads/feature", "success"),
-            ("merge_group", "refs/heads/gh-readonly-queue/release/pr-1", "success"),
-        ):
-            with self.subTest(event=event, ref=ref):
-                expected = expected_results("full", event, ref)
-                self.assertEqual(expected["e2e-full"], required)
-                needs = {job: {"result": result} for job, result in expected.items()}
-                needs["classify-changes"]["outputs"] = {"mode": "full"}
-                for result in ("success", "skipped", "failure", "cancelled"):
-                    needs["e2e-full"]["result"] = result
-                    self.assertEqual(verify_results(needs, event, ref) == [], result == required)
-
     def test_repository_wiring_and_missing_dependency_regression(self):
         self.assertEqual(check_workflow(ROOT), [])
         with tempfile.TemporaryDirectory() as directory: