fix(communication): release the _activeStreams slot when relay ActorOf throws (review 2026-10-21 03/M-2)

src/ZB.MOM.WW.ScadaBridge.Communication/Grpc/SiteStreamGrpcServer.cs

@@ -413,10 +413,28 @@ public class SiteStreamGrpcServer : SiteStreamService.SiteStreamServiceBase
                         correlationId, streamKind, n, channelCapacity);
             });
 
+        // ActorOf can throw when the actor system is terminating through a path that never
+        // set _shuttingDown (the downed-node watchdog). The _activeStreams slot registered
+        // above would otherwise leak against GrpcMaxConcurrentStreams forever (review
+        // 2026-10-21 03/M-2), so release it — and the channel — before rethrowing.
         var actorSeq = Interlocked.Increment(ref _actorCounter);
-        var relayActor = _actorSystem!.ActorOf(
+        IActorRef relayActor;
-            Props.Create(typeof(Actors.StreamRelayActor), correlationId, channel.Writer),
+        try
-            $"stream-relay-{correlationId}-{actorSeq}");
+        {
+            relayActor = _actorSystem!.ActorOf(
+                Props.Create(typeof(Actors.StreamRelayActor), correlationId, channel.Writer),
+                $"stream-relay-{correlationId}-{actorSeq}");
+        }
+        catch (Exception ex)
+        {
+            _logger.LogWarning(ex,
+                "Could not create the relay actor for {Description} (correlation {CorrelationId}); releasing the stream slot.",
+                description, correlationId);
+            channel.Writer.TryComplete();
+            _activeStreams.TryRemove(
+                new KeyValuePair<string, StreamEntry>(correlationId, entry));
+            throw;
+        }
 
         // The previous code called _streamSubscriber.Subscribe
         // OUTSIDE the try block that owns relay-actor cleanup. If Subscribe threw

tests/ZB.MOM.WW.ScadaBridge.Communication.Tests/Grpc/SiteStreamGrpcServerTests.cs

@@ -478,6 +478,29 @@ public class SiteStreamGrpcServerTests : TestKit
         subscriber.DidNotReceive().RemoveSubscriber(Arg.Any<IActorRef>());
     }
 
+    /// <summary>
+    /// Review 2026-10-21 03/M-2: if <c>ActorOf</c> throws (the actor system terminating via
+    /// the downed-node watchdog, so <c>_shuttingDown</c> was never set), the
+    /// <c>_activeStreams</c> slot registered just before it must not leak.
+    /// </summary>
+    /// <returns>A task that represents the asynchronous operation.</returns>
+    [Fact]
+    public async Task ActorOfThrows_DoesNotLeakActiveStreamSlot()
+    {
+        var terminated = ActorSystem.Create("m2-terminated");
+        await terminated.Terminate();
+
+        var server = CreateServer();
+        server.SetReady(terminated);
+
+        var writer = Substitute.For<IServerStreamWriter<SiteStreamEvent>>();
+        await Assert.ThrowsAnyAsync<Exception>(
+            () => server.SubscribeInstance(MakeRequest("corr-m2"), writer, CreateMockContext()));
+
+        Assert.Equal(0, server.ActiveStreamCount);
+        _subscriber.DidNotReceive().Subscribe(Arg.Any<string>(), Arg.Any<IActorRef>());
+    }
+
     /// <summary>Verifies that the active subscribers gauge rises to one on connect and returns to zero on cancellation.</summary>
     /// <returns>A task that represents the asynchronous operation.</returns>
     [Fact]