fix(data-connection-layer): resolve DataConnectionLayer-002/003/004/005 — Resume supervision, concurrent dicts, subscribe-failure classification, write timeout

tests/ScadaLink.DataConnectionLayer.Tests/DataConnectionManagerActorTests.cs

@@ -3,6 +3,7 @@ using Akka.TestKit.Xunit2;
 using NSubstitute;
 using ScadaLink.Commons.Interfaces.Protocol;
 using ScadaLink.Commons.Messages.DataConnection;
+using ScadaLink.Commons.Types.Enums;
 using ScadaLink.DataConnectionLayer.Actors;
 using ScadaLink.HealthMonitoring;
 
@@ -57,6 +58,52 @@ public class DataConnectionManagerActorTests : TestKit
         Assert.Contains("Unknown connection", response.ErrorMessage);
     }
 
+    [Fact]
+    public async Task DCL002_ConnectionActorCrash_PreservesSubscriptionState()
+    {
+        // Regression test for DataConnectionLayer-002. The supervisor used
+        // Directive.Restart, which discards the connection actor's in-memory
+        // subscription registry — breaking the design doc's "transparent
+        // re-subscribe" guarantee (subscribers are never re-subscribed and sit at
+        // stale quality forever). After the fix the supervisor uses Resume, which
+        // keeps the actor instance and its state across a transient exception.
+        var mockAdapter = Substitute.For<IDataConnection>();
+        mockAdapter.ConnectAsync(Arg.Any<IDictionary<string, string>>(), Arg.Any<CancellationToken>())
+            .Returns(Task.CompletedTask);
+        mockAdapter.Status.Returns(ConnectionHealth.Connected);
+        mockAdapter.SubscribeAsync(Arg.Any<string>(), Arg.Any<SubscriptionCallback>(), Arg.Any<CancellationToken>())
+            .Returns("sub-001");
+        mockAdapter.ReadAsync(Arg.Any<string>(), Arg.Any<CancellationToken>())
+            .Returns(new ReadResult(false, null, null));
+        // A write throws synchronously, escaping the message handler and crashing
+        // the connection actor — exercising the supervisor strategy.
+        mockAdapter.WriteAsync(Arg.Any<string>(), Arg.Any<object?>(), Arg.Any<CancellationToken>())
+            .Returns<Task<WriteResult>>(_ => throw new InvalidOperationException("boom"));
+
+        _mockFactory.Create("OpcUa", Arg.Any<IDictionary<string, string>>()).Returns(mockAdapter);
+
+        var manager = Sys.ActorOf(Props.Create(() =>
+            new DataConnectionManagerActor(_mockFactory, _options, _mockHealthCollector)));
+
+        manager.Tell(new CreateConnectionCommand("conn1", "OpcUa", new Dictionary<string, string>(), null, 3));
+        await Task.Delay(300); // connection actor reaches Connected
+
+        // Register a subscription.
+        manager.Tell(new SubscribeTagsRequest("c1", "inst1", "conn1", ["tag1"], DateTimeOffset.UtcNow));
+        ExpectMsg<SubscribeTagsResponse>(TimeSpan.FromSeconds(3));
+
+        // Crash the connection actor via a synchronously-throwing write.
+        manager.Tell(new WriteTagRequest("c2", "conn1", "tag1", 42, DateTimeOffset.UtcNow));
+        await Task.Delay(300); // supervisor handles the failure
+
+        // After the crash the subscription state must survive: the health report
+        // still shows the subscribed/resolved tag. With Restart it would be 0.
+        manager.Tell(new GetAllHealthReports());
+        var report = ExpectMsg<DataConnectionHealthReport>(TimeSpan.FromSeconds(3));
+        Assert.Equal(1, report.TotalSubscribedTags);
+        Assert.Equal(1, report.ResolvedTags);
+    }
+
     [Fact]
     public void CreateConnection_UsesFactory()
     {