fix(mesh-phase5): dial supervisor — re-dial on endpoint change, pin no-restart invariant, throttle skip warnings

Claude-Session: https://claude.ai/code/session_01GASWkNEi68FSCtvr6rLoEW
2026-07-23 16:43:09 -04:00
parent ffb75b7854
commit 84fa2e1e43
3 changed files with 91 additions and 13 deletions
@@ -70,6 +70,44 @@ public sealed class TelemetryDialSupervisorTests : TestKit
        // node-3 gets a fresh dialer.
        var invocations = dial.WaitForCount(3, Timeout);
        invocations.ShouldContain(i => i.Target.NodeId == "node-3" && i.CorrelationId == "central-node-3-1");
+
+        // node-2 is present in BOTH refreshes with an unchanged endpoint: it must NOT be restarted, or
+        // its live stream would be dropped on every refresh. Exactly one dial invocation for it.
+        ExpectNoMsg(TimeSpan.FromMilliseconds(200));
+        dial.Invocations.Count(i => i.Target.NodeId == "node-2").ShouldBe(1);
+    }
+
+    [Fact]
+    public void Refresh_redials_a_node_whose_endpoint_changed_but_not_an_unchanged_one()
+    {
+        var dial = new FakeDialLoop();
+        var current = new List<TelemetryDialTarget>
+        {
+            new("node-1", "http://old-host:5300"),
+            new("node-2", "http://h2:5300"),
+        };
+
+        var actor = Spawn(dial, () => current.ToArray());
+        dial.WaitForCount(2, Timeout);
+        var node1Original = dial.Invocations.Single(i => i.Target.NodeId == "node-1");
+
+        // node-1 re-provisioned to a new host; node-2 unchanged.
+        current.Clear();
+        current.Add(new TelemetryDialTarget("node-1", "http://new-host:5300"));
+        current.Add(new TelemetryDialTarget("node-2", "http://h2:5300"));
+        actor.Tell(new TelemetryDialSupervisor.RefreshNodes());
+
+        // The old node-1 stream is torn down and a fresh dialer opens at the NEW endpoint (gen 2).
+        AwaitCondition(() => node1Original.Ct.IsCancellationRequested, Timeout);
+        var invocations = dial.WaitForCount(3, Timeout);
+        invocations.ShouldContain(i =>
+            i.Target.NodeId == "node-1"
+            && i.Target.Endpoint == "http://new-host:5300"
+            && i.CorrelationId == "central-node-1-2");
+
+        // node-2's endpoint did not change → still exactly one dial invocation (no churn).
+        ExpectNoMsg(TimeSpan.FromMilliseconds(200));
+        dial.Invocations.Count(i => i.Target.NodeId == "node-2").ShouldBe(1);
    }

    [Fact]