fix(communication): site-address load guard cannot wedge on an unexpected fault; lifecycle-cancelled loads log at Debug (review 2026-10-21 03/M-1 follow-up)

2026-10-22 08:29:02 -04:00
parent 4456a04e81
commit e0ebebac11
2 changed files with 69 additions and 7 deletions
@@ -116,4 +116,42 @@ public class CentralCommunicationActorAddressRefreshTests : TestKit
        transport.DidNotReceive().ReconcileSites(
            Arg.Is<SiteAddressCacheLoaded>(c => c.GrpcContacts.ContainsKey("site-old")));
    }
+
+    /// <summary>Verifies that an unexpected Status.Failure releases the overlap guard so later refreshes still load.</summary>
+    [Fact]
+    public void UnexpectedStatusFailure_ReleasesTheOverlapGuard()
+    {
+        var calls = 0;
+        var repo = Substitute.For<ISiteRepository>();
+        repo.GetAllSitesAsync(Arg.Any<CancellationToken>()).Returns(_ =>
+        {
+            Interlocked.Increment(ref calls);
+            return new TaskCompletionSource<IReadOnlyList<Site>>().Task; // never completes
+        });
+
+        var (actor, _) = CreateActor(repo);
+        AwaitCondition(() => Volatile.Read(ref calls) == 1, TimeSpan.FromSeconds(3));
+
+        actor.Tell(new Status.Failure(new InvalidOperationException("unexpected pipe fault")));
+        actor.Tell(new RefreshSiteAddresses());
+
+        AwaitCondition(() => Volatile.Read(ref calls) == 2, TimeSpan.FromSeconds(3));
+    }
+
+    /// <summary>Verifies that a load cancelled by the actor lifecycle (stop/restart) is not logged as a Warning.</summary>
+    [Fact]
+    public void LoadCancelledByActorLifecycle_IsNotAWarning()
+    {
+        var repo = Substitute.For<ISiteRepository>();
+        repo.GetAllSitesAsync(Arg.Any<CancellationToken>()).Returns(new List<Site>());
+        var (actor, _) = CreateActor(repo);
+
+        // A previous incarnation's load, cancelled by its PostStop, lands on the restarted actor.
+        EventFilter.Warning(contains: "site addresses").Expect(0, TimeSpan.FromSeconds(1), () =>
+            actor.Tell(new SiteAddressLoadFailed(1, new OperationCanceledException(), null, ActorStopping: true)));
+
+        // A genuine DB failure still warns.
+        EventFilter.Warning(contains: "site addresses").ExpectOne(TimeSpan.FromSeconds(3), () =>
+            actor.Tell(new SiteAddressLoadFailed(2, new InvalidOperationException("db down"), null)));
+    }
 }