fix(r2-10): in-process resilience status store

archreview/plans/R2-10-resilience-observability-plan.md.tasks.json

@@ -8,7 +8,7 @@
     { "id": "T4", "subject": "Commons: DriverResilienceStatusChanged record + driver-resilience-status TopicName", "status": "completed", "blockedBy": [] },
     { "id": "T5", "subject": "Host: DriverResilienceStatusPublisherService (BuildMessages mapping + 5s DPS timer shell; TDD mapping in Host.IntegrationTests)", "status": "completed", "blockedBy": ["T1", "T4"] },
     { "id": "T6", "subject": "Host DI: register the publisher in AddOtOpcUaDriverFactories + ResilienceStatusReaderRegistrationTests wiring guard (tracker HAS a production reader)", "status": "completed", "blockedBy": ["T5"] },
-    { "id": "T7", "subject": "AdminUI: IDriverResilienceStatusStore + InMemory impl + AddOtOpcUaDriverStatusServices registration (TDD mirroring DriverStatusSnapshotStoreTests)", "status": "pending", "blockedBy": ["T4"] },
+    { "id": "T7", "subject": "AdminUI: IDriverResilienceStatusStore + InMemory impl + AddOtOpcUaDriverStatusServices registration (TDD mirroring DriverStatusSnapshotStoreTests)", "status": "completed", "blockedBy": ["T4"] },
     { "id": "T8", "subject": "AdminUI: DriverResilienceStatusBridge DPS actor + spawn in WithOtOpcUaSignalRBridges", "status": "pending", "blockedBy": ["T7"] },
     { "id": "T9", "subject": "AdminUI: resilience chip section in DriverStatusPanel.razor (in-process store read; no bUnit — verified by T10)", "status": "pending", "blockedBy": ["T7"] },
     { "id": "T10", "subject": "Live-/run gate on docker-dev: rebuild BOTH centrals, S7 dead-endpoint breaker-open, both-replica chips, recovery clear, staleness dim", "status": "pending", "blockedBy": ["T6", "T8", "T9"] },

src/Server/ZB.MOM.WW.OtOpcUa.AdminUI/Hubs/HubServiceCollectionExtensions.cs

@@ -28,6 +28,7 @@ public static class HubServiceCollectionExtensions
     public static IServiceCollection AddOtOpcUaDriverStatusServices(this IServiceCollection services)
     {
         services.AddSingleton<IDriverStatusSnapshotStore, InMemoryDriverStatusSnapshotStore>();
+        services.AddSingleton<IDriverResilienceStatusStore, InMemoryDriverResilienceStatusStore>();
         services.AddSingleton(typeof(IInProcessBroadcaster<>), typeof(InProcessBroadcaster<>));
         return services;
     }

src/Server/ZB.MOM.WW.OtOpcUa.AdminUI/Hubs/IDriverResilienceStatusStore.cs

@@ -0,0 +1,36 @@
+using ZB.MOM.WW.OtOpcUa.Commons.Messages.Drivers;
+
+namespace ZB.MOM.WW.OtOpcUa.AdminUI.Hubs;
+
+/// <summary>
+///     Singleton last-snapshot-per-<c>(instance, host)</c> cache of driver resilience status
+///     (breaker open / consecutive failures / in-flight depth). Populated by
+///     <c>DriverResilienceStatusBridge</c> as it forwards <c>driver-resilience-status</c> DPS
+///     messages; read in-process by the driver status panel's resilience section via
+///     <see cref="SnapshotChanged"/> + <see cref="GetForInstance"/>. Mirrors
+///     <see cref="IDriverStatusSnapshotStore"/> — the panel reads this singleton directly instead of
+///     opening a self-targeted SignalR connection (which a server-side Blazor component cannot reach
+///     behind a reverse proxy).
+/// </summary>
+public interface IDriverResilienceStatusStore
+{
+    /// <summary>Stores or replaces the last-known resilience snapshot for a <c>(instance, host)</c> pair.</summary>
+    /// <param name="snapshot">The resilience status snapshot to store.</param>
+    void Upsert(DriverResilienceStatusChanged snapshot);
+
+    /// <summary>Returns every stored host snapshot for one driver instance (empty if none).</summary>
+    /// <param name="driverInstanceId">The driver instance's stable logical ID.</param>
+    /// <returns>The per-host resilience snapshots for that instance.</returns>
+    IReadOnlyList<DriverResilienceStatusChanged> GetForInstance(string driverInstanceId);
+
+    /// <summary>Returns a point-in-time snapshot of every tracked <c>(instance, host)</c> pair.</summary>
+    /// <returns>A read-only collection of the last-known resilience snapshot for each pair.</returns>
+    IReadOnlyCollection<DriverResilienceStatusChanged> GetAll();
+
+    /// <summary>
+    ///     Raised after every <see cref="Upsert"/> with the just-stored snapshot. Lets the Blazor Server
+    ///     driver status panel receive live updates by reading this singleton directly. Handlers run on
+    ///     the caller's thread (the bridge actor), so subscribers must marshal to their own sync context.
+    /// </summary>
+    event Action<DriverResilienceStatusChanged>? SnapshotChanged;
+}

src/Server/ZB.MOM.WW.OtOpcUa.AdminUI/Hubs/InMemoryDriverResilienceStatusStore.cs

@@ -0,0 +1,33 @@
+using System.Collections.Concurrent;
+using ZB.MOM.WW.OtOpcUa.Commons.Messages.Drivers;
+
+namespace ZB.MOM.WW.OtOpcUa.AdminUI.Hubs;
+
+/// <summary>
+///     Thread-safe in-memory implementation of <see cref="IDriverResilienceStatusStore"/>.
+///     Keyed by <c>(DriverInstanceId, HostName)</c>; last write wins.
+/// </summary>
+public sealed class InMemoryDriverResilienceStatusStore : IDriverResilienceStatusStore
+{
+    private readonly ConcurrentDictionary<(string Instance, string Host), DriverResilienceStatusChanged> _byPair = new();
+
+    /// <inheritdoc />
+    public event Action<DriverResilienceStatusChanged>? SnapshotChanged;
+
+    /// <inheritdoc />
+    public void Upsert(DriverResilienceStatusChanged snapshot)
+    {
+        _byPair[(snapshot.DriverInstanceId, snapshot.HostName)] = snapshot;
+        // Capture-then-invoke so a concurrent unsubscribe can't null the delegate mid-raise.
+        SnapshotChanged?.Invoke(snapshot);
+    }
+
+    /// <inheritdoc />
+    public IReadOnlyList<DriverResilienceStatusChanged> GetForInstance(string driverInstanceId) =>
+        _byPair.Values
+            .Where(s => string.Equals(s.DriverInstanceId, driverInstanceId, StringComparison.Ordinal))
+            .ToList();
+
+    /// <inheritdoc />
+    public IReadOnlyCollection<DriverResilienceStatusChanged> GetAll() => _byPair.Values.ToArray();
+}

tests/Server/ZB.MOM.WW.OtOpcUa.AdminUI.Tests/DriverResilienceStatusStoreTests.cs

@@ -0,0 +1,87 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.AdminUI.Hubs;
+using ZB.MOM.WW.OtOpcUa.Commons.Messages.Drivers;
+
+namespace ZB.MOM.WW.OtOpcUa.AdminUI.Tests;
+
+/// <summary>
+///     Covers the in-process push contract the Blazor Server driver status panel's resilience section
+///     relies on: <see cref="IDriverResilienceStatusStore.SnapshotChanged"/> fires on every
+///     <see cref="IDriverResilienceStatusStore.Upsert"/>, keyed per <c>(instance, host)</c> with
+///     last-write-wins, and <c>GetForInstance</c> returns only the queried instance's hosts. Mirrors
+///     <see cref="DriverStatusSnapshotStoreTests"/>; the panel reads this store directly (no
+///     self-targeted SignalR connection).
+/// </summary>
+public sealed class DriverResilienceStatusStoreTests
+{
+    private static DriverResilienceStatusChanged Msg(
+        string instance, string host, bool breakerOpen = false, int failures = 0) =>
+        new(instance, host, breakerOpen, failures, 0, null,
+            new DateTime(2026, 7, 13, 0, 0, 0, DateTimeKind.Utc),
+            new DateTime(2026, 7, 13, 0, 0, 0, DateTimeKind.Utc));
+
+    [Fact]
+    public void Upsert_raises_SnapshotChanged_with_the_stored_snapshot()
+    {
+        var store = new InMemoryDriverResilienceStatusStore();
+        var received = new List<DriverResilienceStatusChanged>();
+        store.SnapshotChanged += received.Add;
+
+        var msg = Msg("drv-1", "host-a", breakerOpen: true);
+        store.Upsert(msg);
+
+        received.Count.ShouldBe(1);
+        received[0].ShouldBeSameAs(msg);
+    }
+
+    [Fact]
+    public void Upsert_is_last_write_wins_per_instance_host()
+    {
+        var store = new InMemoryDriverResilienceStatusStore();
+        store.Upsert(Msg("drv-1", "host-a", failures: 3));
+        store.Upsert(Msg("drv-1", "host-a", failures: 0));
+
+        var hosts = store.GetForInstance("drv-1");
+        hosts.Count.ShouldBe(1);
+        hosts[0].ConsecutiveFailures.ShouldBe(0);
+    }
+
+    [Fact]
+    public void GetForInstance_returns_only_the_queried_instances_hosts()
+    {
+        var store = new InMemoryDriverResilienceStatusStore();
+        store.Upsert(Msg("drv-1", "host-a"));
+        store.Upsert(Msg("drv-1", "host-b"));
+        store.Upsert(Msg("drv-2", "host-a"));
+
+        var hosts = store.GetForInstance("drv-1");
+
+        hosts.Count.ShouldBe(2);
+        hosts.ShouldAllBe(h => h.DriverInstanceId == "drv-1");
+    }
+
+    [Fact]
+    public void GetForInstance_returns_empty_for_unknown_instance()
+    {
+        var store = new InMemoryDriverResilienceStatusStore();
+        store.Upsert(Msg("drv-1", "host-a"));
+
+        store.GetForInstance("nope").ShouldBeEmpty();
+    }
+
+    [Fact]
+    public void Unsubscribed_handler_stops_receiving_after_removal()
+    {
+        var store = new InMemoryDriverResilienceStatusStore();
+        var count = 0;
+        void Handler(DriverResilienceStatusChanged _) => count++;
+
+        store.SnapshotChanged += Handler;
+        store.Upsert(Msg("drv-1", "host-a"));
+        store.SnapshotChanged -= Handler;
+        store.Upsert(Msg("drv-1", "host-a"));
+
+        count.ShouldBe(1);
+    }
+}