perf(comms+audit): close phase-2 residuals — direct ingest path, monotonic timeouts, synthetic probe, not-reporting set, cursor-exact audit pull

2026-08-14 21:38:23 -04:00
parent 4cd1441984
commit a5882753dd
38 changed files with 1254 additions and 443 deletions
@@ -81,7 +81,7 @@ public class SiteAlarmLiveCacheServiceTests : TestKit
    }

    private SiteAlarmLiveCacheService CreateService(TimeSpan linger, out CountingFactory factory,
-        int maxSubscribersPerSite = 200)
+        int maxSubscribersPerSite = 200, IReadOnlyList<Instance>? enabledInstances = null)
    {
        // Site with gRPC addresses, and NO enabled instances → the seed fan-out returns
        // empty immediately (so IsLive flips true fast without any snapshot Asks).
@@ -97,7 +97,7 @@ public class SiteAlarmLiveCacheServiceTests : TestKit

        var instanceRepo = Substitute.For<ITemplateEngineRepository>();
        instanceRepo.GetInstancesBySiteIdAsync(SiteId, Arg.Any<CancellationToken>())
-            .Returns(new List<Instance>());
+            .Returns((IReadOnlyList<Instance>)(enabledInstances ?? new List<Instance>()));

        var services = new ServiceCollection();
        services.AddScoped(_ => siteRepo);
@@ -121,6 +121,45 @@ public class SiteAlarmLiveCacheServiceTests : TestKit
        return service;
    }

+    [Fact]
+    public void Seed_FanOut_Publishes_The_Instances_That_Failed_To_Answer()
+    {
+        // Arch-review phase-2 residual #4: the seed/reconcile fan-out already knows which
+        // Enabled instances failed to answer and used to discard it, forcing the Alarm Summary
+        // page to run a SECOND identical fan-out purely to rebuild that list. It is now
+        // published alongside the snapshot.
+        var instances = new List<Instance>
+        {
+            new("inst-b") { Id = 2, SiteId = SiteId, State = InstanceState.Enabled },
+            new("inst-a") { Id = 1, SiteId = SiteId, State = InstanceState.Enabled },
+            // Disabled instances are never fanned out, so they can never be "not reporting".
+            new("inst-off") { Id = 3, SiteId = SiteId, State = InstanceState.Disabled },
+        };
+
+        // The CommunicationService has no site actor wired, so every snapshot Ask faults —
+        // which is exactly the "instance did not answer" case.
+        var service = CreateService(TimeSpan.FromMilliseconds(200), out _, enabledInstances: instances);
+
+        using var sub = service.Subscribe(SiteId, () => { });
+        AwaitCondition(() => service.IsLive(SiteId), TimeSpan.FromSeconds(5));
+
+        AwaitCondition(
+            () => service.GetNotReportingInstances(SiteId).Count == 2,
+            TimeSpan.FromSeconds(5));
+
+        // Ordered by name (ordinal-ignore-case), matching AlarmSummaryService's poll output so
+        // the page renders identically whichever source supplied the list.
+        Assert.Equal(new[] { "inst-a", "inst-b" }, service.GetNotReportingInstances(SiteId));
+    }
+
+    [Fact]
+    public void NotReporting_Is_Empty_For_An_Unknown_Site()
+    {
+        var service = CreateService(TimeSpan.FromMilliseconds(200), out _);
+
+        Assert.Empty(service.GetNotReportingInstances(SiteId));
+    }
+
    [Fact]
    public void First_Subscriber_Starts_One_Aggregator_Shared_By_Multiple_Viewers()
    {