Merge lane A: runtime driver lifecycle (#731)

Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
This commit is contained in:
Joseph Doherty
2026-10-22 10:44:22 -04:00
24 changed files with 2344 additions and 65 deletions
@@ -0,0 +1,153 @@
using System.Collections.Concurrent;
using Akka.Actor;
using Akka.TestKit.Xunit2;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.AdminUI.Hubs;
using ZB.MOM.WW.OtOpcUa.AdminUI.Telemetry;
using ZB.MOM.WW.OtOpcUa.Cluster;
using ZB.MOM.WW.OtOpcUa.Commons.Messages.Alerts;
using ZB.MOM.WW.OtOpcUa.Commons.Messages.Drivers;
using ZB.MOM.WW.OtOpcUa.Commons.Messages.Logging;
namespace ZB.MOM.WW.OtOpcUa.AdminUI.Tests.Telemetry;
/// <summary>
/// Review 02 U2: <c>TelemetryDial:CallTimeoutSeconds</c> was bound, defaulted and documented, and consumed
/// nowhere. It now bounds a dial's <b>establishment</b>: a dial that has delivered no event by the deadline is
/// cut and re-dialled with a failure streak. A live node answers a fresh subscribe at once (the node-local hub
/// replays its health/resilience snapshots), so the deadline bites only a dial that never connects — a
/// black-holed host, where the underlying socket connect has no timeout of its own. An established stream
/// is deliberately NOT bounded: HTTP/2 keepalive pings own its liveness, and the wire has no heartbeat, so an
/// idle timeout would churn every quiet node.
/// </summary>
public sealed class TelemetryDialDeadlineTests : TestKit
{
private static readonly TimeSpan Timeout = TimeSpan.FromSeconds(8);
[Fact]
public void A_dial_that_delivers_no_event_by_the_deadline_is_cancelled_and_redialled()
{
var dial = new FakeDialLoop();
Spawn(dial, callTimeoutSeconds: 1);
var first = dial.WaitForCount(1, Timeout).Single();
first.Ct.IsCancellationRequested.ShouldBeFalse("the deadline has not elapsed yet");
// The stalled dial is cut at the deadline...
AwaitCondition(() => first.Ct.IsCancellationRequested, Timeout);
// ...and the supervisor re-dials (a stalled dial is a transport failure, never a silent end).
var invocations = dial.WaitForCount(2, Timeout);
invocations[1].CorrelationId.ShouldBe("central-node-1-2");
}
[Fact]
public async Task A_dial_that_delivers_an_event_before_the_deadline_is_left_open()
{
var dial = new FakeDialLoop { DeliverOnOpen = true };
var sinks = new Sinks();
Spawn(dial, callTimeoutSeconds: 1, sinks);
var first = dial.WaitForCount(1, Timeout).Single();
AwaitCondition(() => sinks.HealthStore.Upserts.Count == 1, Timeout);
// Positive control: well past the deadline, the established stream is untouched.
await Task.Delay(TimeSpan.FromSeconds(2.5));
first.Ct.IsCancellationRequested.ShouldBeFalse("an established stream must never be cut by the dial deadline");
dial.Invocations.Count.ShouldBe(1);
}
private IActorRef Spawn(FakeDialLoop dial, int callTimeoutSeconds, Sinks? sinks = null)
{
sinks ??= new Sinks();
var options = new TelemetryDialOptions { ContactRefreshSeconds = 3600, CallTimeoutSeconds = callTimeoutSeconds };
return Sys.ActorOf(TelemetryDialSupervisor.Props(
() => Task.FromResult<IReadOnlyList<TelemetryDialTarget>>(
new[] { new TelemetryDialTarget("node-1", "http://h1:5300") }),
dial.Loop,
sinks.AlarmBroadcaster,
sinks.ScriptBroadcaster,
sinks.HealthStore,
sinks.ResilienceStore,
options));
}
private static DriverHealthChanged Health(string id) =>
new("C1", id, "Healthy", DateTime.UtcNow, null, 0, DateTime.UtcNow);
private sealed class DialInvocation
{
private readonly TaskCompletionSource _completion = new(TaskCreationOptions.RunContinuationsAsynchronously);
public required string CorrelationId { get; init; }
public required CancellationToken Ct { get; init; }
public Task Completion => _completion.Task;
public void Complete() => _completion.TrySetResult();
}
/// <summary>Records each invocation; optionally delivers one health event on open; ends only when cancelled.</summary>
private sealed class FakeDialLoop
{
private readonly ConcurrentQueue<DialInvocation> _invocations = new();
public bool DeliverOnOpen { get; init; }
public IReadOnlyList<DialInvocation> Invocations => _invocations.ToArray();
public TelemetryDialLoop Loop => async (_, correlationId, onMapped, _, ct) =>
{
var invocation = new DialInvocation { CorrelationId = correlationId, Ct = ct };
_invocations.Enqueue(invocation);
if (DeliverOnOpen) onMapped(Health("drv-1"));
await using var reg = ct.Register(() => invocation.Complete());
await invocation.Completion;
};
public IReadOnlyList<DialInvocation> WaitForCount(int count, TimeSpan timeout)
{
var deadline = DateTime.UtcNow + timeout;
while (DateTime.UtcNow < deadline)
{
var snapshot = _invocations.ToArray();
if (snapshot.Length >= count) return snapshot;
Thread.Sleep(10);
}
throw new Xunit.Sdk.XunitException(
$"Expected at least {count} dial invocations within {timeout}, saw {_invocations.Count}.");
}
}
private sealed class Sinks
{
public NullBroadcaster<AlarmTransitionEvent> AlarmBroadcaster { get; } = new();
public NullBroadcaster<ScriptLogEntry> ScriptBroadcaster { get; } = new();
public RecordingHealthStore HealthStore { get; } = new();
public NullResilienceStore ResilienceStore { get; } = new();
}
private sealed class NullBroadcaster<T> : IInProcessBroadcaster<T>
{
public event Action<T>? Received;
public event Action<bool>? ConnectionStateChanged;
public bool IsConnected { get; private set; }
public void Publish(T item) => Received?.Invoke(item);
public void SetConnected(bool connected) { IsConnected = connected; ConnectionStateChanged?.Invoke(connected); }
}
private sealed class RecordingHealthStore : IDriverStatusSnapshotStore
{
public ConcurrentQueue<DriverHealthChanged> Upserts { get; } = new();
public event Action<DriverHealthChanged>? SnapshotChanged;
public void Upsert(DriverHealthChanged snapshot) { Upserts.Enqueue(snapshot); SnapshotChanged?.Invoke(snapshot); }
public bool TryGet(string driverInstanceId, out DriverHealthChanged snapshot) { snapshot = null!; return false; }
public IReadOnlyCollection<DriverHealthChanged> GetAll() => Upserts.ToArray();
}
private sealed class NullResilienceStore : IDriverResilienceStatusStore
{
public event Action<DriverResilienceStatusChanged>? SnapshotChanged;
public void Upsert(DriverResilienceStatusChanged snapshot) => SnapshotChanged?.Invoke(snapshot);
public IReadOnlyList<DriverResilienceStatusChanged> GetForInstance(string driverInstanceId) => Array.Empty<DriverResilienceStatusChanged>();
public IReadOnlyCollection<DriverResilienceStatusChanged> GetAll() => Array.Empty<DriverResilienceStatusChanged>();
}
}
@@ -64,7 +64,7 @@ public sealed class CalculationDependencyFlowTests : RuntimeActorTestBase
{
published.Enqueue(e);
var quality = e.Snapshot.StatusCode == 0u ? OpcUaStatus.Good : OpcUaStatus.Bad;
mux.Tell(new DriverInstanceActor.AttributeValuePublished(
mux.Tell(ValuePublishes.Single(
"calc-drv", e.FullReference, e.Snapshot.Value, quality, Ts));
};
@@ -75,7 +75,7 @@ public sealed class CalculationDependencyFlowTests : RuntimeActorTestBase
// compute dedups) until the computed value has flowed all the way through.
AwaitAssert(() =>
{
mux.Tell(new DriverInstanceActor.AttributeValuePublished(
mux.Tell(ValuePublishes.Single(
"src-drv", "src/a", 21.0, OpcUaStatus.Good, Ts));
// Leg 1 — dep flow: src/a=21 → A computes 42 and publishes on calc/engine/out.
@@ -118,14 +118,14 @@ public sealed class CalculationDependencyFlowTests : RuntimeActorTestBase
{
published.Enqueue(e);
var quality = e.Snapshot.StatusCode == 0u ? OpcUaStatus.Good : OpcUaStatus.Bad;
mux.Tell(new DriverInstanceActor.AttributeValuePublished("calc-drv", e.FullReference, e.Snapshot.Value, quality, Ts));
mux.Tell(ValuePublishes.Single("calc-drv", e.FullReference, e.Snapshot.Value, quality, Ts));
};
var adapter = Sys.ActorOf(DependencyConsumerMuxAdapter.Props(driver, mux), "dep-adapter2");
// Leg 1 — before redeploy: src/a=2 → keep=20, remove=7.
AwaitAssert(() =>
{
mux.Tell(new DriverInstanceActor.AttributeValuePublished("src-drv", "src/a", 2.0, OpcUaStatus.Good, Ts));
mux.Tell(ValuePublishes.Single("src-drv", "src/a", 2.0, OpcUaStatus.Good, Ts));
published.ShouldContain(e => e.FullReference == "calc/keep" && Equals(e.Snapshot.Value, 20.0));
published.ShouldContain(e => e.FullReference == "calc/remove" && Equals(e.Snapshot.Value, 7.0));
}, PresenceBudget, TimeSpan.FromMilliseconds(100));
@@ -150,8 +150,8 @@ public sealed class CalculationDependencyFlowTests : RuntimeActorTestBase
// new result; the REMOVED tag never publishes a value derived from the post-redeploy input.
AwaitAssert(() =>
{
mux.Tell(new DriverInstanceActor.AttributeValuePublished("src-drv", "src/b", 50.0, OpcUaStatus.Good, Ts));
mux.Tell(new DriverInstanceActor.AttributeValuePublished("src-drv", "src/a", 3.0, OpcUaStatus.Good, Ts));
mux.Tell(ValuePublishes.Single("src-drv", "src/b", 50.0, OpcUaStatus.Good, Ts));
mux.Tell(ValuePublishes.Single("src-drv", "src/a", 3.0, OpcUaStatus.Good, Ts));
// (a) new dep B flows through the RE-REGISTERED interest → new tag computes: 50 + 100 = 150.
published.ShouldContain(e => e.FullReference == "calc/add" && Equals(e.Snapshot.Value, 150.0));
@@ -0,0 +1,79 @@
using Akka.Actor;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.Interfaces;
using ZB.MOM.WW.OtOpcUa.Commons.Messages.Fleet;
using ZB.MOM.WW.OtOpcUa.Commons.Types;
using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
/// <summary>
/// Review 02 U1: <c>GetDiagnostics</c> answered a placeholder — <c>Guid.Empty</c>, "Spawned", zero devices and
/// "now" for every child, in every behaviour — so the snapshot looked live and said nothing. It now carries
/// each child's row id, instance id, last-published health state, device counts and last change time.
/// </summary>
public sealed class DriverHostActorDiagnosticsTests : RuntimeActorTestBase
{
private static readonly RevisionHash RevA = RevisionHash.Parse(new string('a', 64));
private static readonly NodeId TestNode = NodeId.Parse("node-1:4053");
private const string DriverId = "DI-1";
[Fact]
public async Task A_connected_child_reports_its_row_id_name_health_state_and_device_count()
{
var db = NewInMemoryDbFactory();
var rowId = Guid.NewGuid();
var before = DateTime.UtcNow.AddSeconds(-1);
var factory = new RecordingSubscribableDriverFactory("Modbus");
var deploymentId = V3ArtifactSeeder.SeedV3Deployment(
db, RevA, new[] { (DriverId, "Line1-Modbus", "dev1", "speed", "Float", false) },
Array.Empty<(string, string, string, string, string?)>(),
enabled: true, driverRowIds: new Dictionary<string, Guid> { [DriverId] = rowId });
var (host, _, _) = V3ArtifactSeeder.SpawnHostAndApply(this, db, TestNode, deploymentId, RevA, factory);
var snapshot = await AwaitDiagnosticsAsync(host, s => s.Drivers.Any(d => d.State == "Healthy"));
snapshot.NodeId.ShouldBe(TestNode);
snapshot.CurrentRevision.ShouldBe(RevA);
snapshot.RunningFromCache.ShouldBeFalse();
var driver = snapshot.Drivers.ShouldHaveSingleItem();
driver.DriverInstanceId.ShouldBe(rowId, "the real row id, not Guid.Empty");
driver.Name.ShouldBe(DriverId, "the string instance id every other surface keys on");
driver.State.ShouldBe("Healthy");
driver.ConnectedDevices.ShouldBe(1);
driver.FaultedDevices.ShouldBe(0);
driver.LastChangeUtc.ShouldBeGreaterThan(before);
driver.LastChangeUtc.ShouldBeLessThanOrEqualTo(DateTime.UtcNow);
}
[Fact]
public async Task A_stubbed_child_reports_Stubbed_with_no_devices()
{
var db = NewInMemoryDbFactory();
// The factory answers for a type the artifact does not use, so the host stubs the child.
var factory = new RecordingSubscribableDriverFactory("NotModbus");
var deploymentId = V3ArtifactSeeder.SeedV3Deployment(
db, RevA, new[] { (DriverId, "Modbus", "dev1", "speed", "Float", false) },
Array.Empty<(string, string, string, string, string?)>(), enabled: true);
var (host, _, _) = V3ArtifactSeeder.SpawnHostAndApply(this, db, TestNode, deploymentId, RevA, factory);
var snapshot = await AwaitDiagnosticsAsync(host, s => s.Drivers.Count == 1);
var driver = snapshot.Drivers.Single();
driver.State.ShouldBe("Stubbed");
driver.ConnectedDevices.ShouldBe(0);
driver.FaultedDevices.ShouldBe(0);
}
private async Task<NodeDiagnosticsSnapshot> AwaitDiagnosticsAsync(IActorRef host, Func<NodeDiagnosticsSnapshot, bool> ready)
{
NodeDiagnosticsSnapshot? snapshot = null;
await AwaitAssertAsync(async () =>
{
snapshot = await host.Ask<NodeDiagnosticsSnapshot>(new GetDiagnostics(CorrelationId.NewId()), PresenceBudget);
ready(snapshot).ShouldBeTrue();
}, PresenceBudget);
return snapshot!;
}
}
@@ -0,0 +1,63 @@
using Akka.Actor;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.Types;
using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
/// <summary>
/// Review 02 L2: <c>DriverHostActor._lastReprovisionUtc</c> was never pruned — a driver removed by a deploy
/// after a supervision crash kept its stamp, so the map grew with every id ever reprovisioned-then-removed,
/// and (the observable half) a driver removed and later re-added inherited the OLD stamp: its first crash
/// after the re-add was deferred by the backoff as if it were crash-looping, although the new incarnation had
/// never crashed. <c>StopChild</c> now drops the stamp.
/// </summary>
public sealed class DriverHostActorReprovisionStampTests : RuntimeActorTestBase
{
private static readonly RevisionHash RevA = RevisionHash.Parse(new string('a', 64));
private static readonly RevisionHash RevB = RevisionHash.Parse(new string('b', 64));
private static readonly RevisionHash RevC = RevisionHash.Parse(new string('c', 64));
private static readonly NodeId TestNode = NodeId.Parse("node-1:4053");
private const string DriverId = "DI-1";
private const string SpeedPath = "Plant/Modbus/dev1/speed";
[Fact]
public void A_driver_removed_and_re_added_does_not_inherit_the_old_incarnations_reprovision_stamp()
{
var db = NewInMemoryDbFactory();
// Far longer than the presence budget below: an inherited stamp would defer the re-added driver's
// reprovision past it, which is the failure this test reports.
var backoff = TimeSpan.FromSeconds(60);
var factory = new RecordingSubscribableDriverFactory("Modbus");
var withDriver = new[] { (DriverId, "Modbus", "dev1", "speed", "Float", false) };
var none = Array.Empty<(string, string, string, string, string?)>();
var revA = V3ArtifactSeeder.SeedV3Deployment(db, RevA, withDriver, none, enabled: true);
var (host, _, _, coordinator) = V3ArtifactSeeder.SpawnHostAndApplyWithCoordinator(
this, db, TestNode, revA, RevA, factory, reprovisionBackoff: backoff);
AwaitAssert(() => factory.Created.Count.ShouldBe(1), PresenceBudget);
AwaitAssert(() => factory.Created[0].SubscribeCount.ShouldBe(1), PresenceBudget);
// Crash 1 on the first incarnation: reprovisioned immediately, which stamps the id.
factory.Created[0].FireMalformedDataChange(SpeedPath);
AwaitAssert(() => factory.Created[0].InitializeCount.ShouldBe(2), PresenceBudget);
// Deploy B removes the driver (StopChild), deploy C re-adds it: a brand-new driver object.
var revB = V3ArtifactSeeder.SeedV3Deployment(
db, RevB, Array.Empty<(string, string, string, string, string, bool)>(), none, enabled: true);
V3ArtifactSeeder.Dispatch(host, coordinator, revB, RevB);
var revC = V3ArtifactSeeder.SeedV3Deployment(db, RevC, withDriver, none, enabled: true);
V3ArtifactSeeder.Dispatch(host, coordinator, revC, RevC);
AwaitAssert(() => factory.Created.Count.ShouldBe(2), PresenceBudget);
var reAdded = factory.Created[1];
AwaitAssert(() => reAdded.SubscribeCount.ShouldBe(1), PresenceBudget);
// Its FIRST crash must reprovision immediately — it has no history of its own.
reAdded.FireMalformedDataChange(SpeedPath);
AwaitAssert(
() => reAdded.InitializeCount.ShouldBe(
2, "the re-added driver's first reprovision was deferred by the REMOVED incarnation's stamp"),
duration: TimeSpan.FromSeconds(5));
}
}
@@ -0,0 +1,66 @@
using Akka.Actor;
using Akka.Event;
using Microsoft.EntityFrameworkCore;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.OpcUa;
using ZB.MOM.WW.OtOpcUa.Commons.Types;
using ZB.MOM.WW.OtOpcUa.Configuration;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
/// <summary>
/// Review 02 H5: the <c>Stale</c> behaviour registered no receive for a driver child's value/alarm bursts (nor
/// for a late <c>ArtifactForApply</c> / <c>ApplyWedged</c> pipe-back), so any that landed there hit
/// <c>Unhandled</c> — dead-letter noise and a dropped message with no log naming why. They are now explicit
/// drops. The two private pipe-back records cannot be constructed from a test; the two public child
/// messages are pinned here, and the private pair by the same registration pattern (compile-time).
/// </summary>
public sealed class DriverHostActorStaleDropTests : RuntimeActorTestBase
{
private static readonly NodeId TestNode = NodeId.Parse("stale-drop-test");
[Fact]
public void Child_value_and_alarm_bursts_landing_in_Stale_are_dropped_not_unhandled()
{
var unhandled = CreateTestProbe();
Sys.EventStream.Subscribe(unhandled.Ref, typeof(UnhandledMessage));
// ThrowingDbFactory ⇒ Bootstrap's catch ⇒ Become(Stale).
var host = Sys.ActorOf(DriverHostActor.Props(
new ThrowingDbFactory(), TestNode, CreateTestProbe().Ref,
localRoles: new HashSet<string> { "driver" }));
// Barrier: Stale answers RouteNodeWrite synchronously, so a reply proves the host is in Stale.
var asker = CreateTestProbe();
host.Tell(new DriverHostActor.RouteNodeWrite("eq-x/speed", 0.0, AddressSpaceRealm.Uns), asker.Ref);
asker.ExpectMsg<DriverHostActor.NodeWriteResult>(PresenceBudget).Reason!.ShouldContain("stale");
// Positive control: an unknown message IS reported as unhandled by this host.
host.Tell("not a message the host knows");
unhandled.ExpectMsg<UnhandledMessage>(m => ReferenceEquals(m.Recipient, host) || m.Recipient.Equals(host), PresenceBudget);
host.Tell(new DriverInstanceActor.AttributeValueBatchPublished(
"DI-1", new[] { new DriverInstanceActor.AttributeValueItem("Plant/x/y", 1.0, OpcUaStatus.Good, DateTime.UtcNow) }));
host.Tell(new DriverInstanceActor.AttributeAlarmPublished(
"DI-1", new AlarmEventArgs(new Handle(), "Plant/x/alarm", "Plant/x/alarm", "AlarmCondition", "test",
AlarmSeverity.High, DateTime.UtcNow)));
// Barrier again, then the absence window.
host.Tell(new DriverHostActor.RouteNodeWrite("eq-x/speed", 0.0, AddressSpaceRealm.Uns), asker.Ref);
asker.ExpectMsg<DriverHostActor.NodeWriteResult>(PresenceBudget);
unhandled.ExpectNoMsg(TimeSpan.FromMilliseconds(500));
}
private sealed class Handle : IAlarmSubscriptionHandle
{
public string DiagnosticId => "stale-test";
}
private sealed class ThrowingDbFactory : IDbContextFactory<OtOpcUaConfigDbContext>
{
public OtOpcUaConfigDbContext CreateDbContext() => throw new InvalidOperationException("db unreachable");
}
}
@@ -0,0 +1,288 @@
using System.Collections.Concurrent;
using System.Diagnostics;
using Akka.Actor;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.OpcUa;
using ZB.MOM.WW.OtOpcUa.Commons.Types;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
/// <summary>
/// Review 02 P2: <c>WriteAttribute</c> was a <c>ReceiveAsync</c>, which suspends the mailbox for the whole
/// device round-trip (up to the 5 s bound). Writes therefore serialised — the SECOND of two operator writes to
/// a ~3 s device missed the host's 8 s Ask and was reverted although the device took it — and every data
/// change queued behind the write. Writes now run off the mailbox under a per-driver in-flight cap; the
/// subscribe/unsubscribe pair that must serialise against the subscription handle stays <c>ReceiveAsync</c>.
/// </summary>
public sealed class DriverInstanceActorConcurrentWriteTests : RuntimeActorTestBase
{
private static readonly TimeSpan WriteLatency = TimeSpan.FromMilliseconds(1200);
private (IActorRef Actor, Akka.TestKit.TestProbe Parent) Connect(
IDriver driver, int? maxWritesInFlight = null, TimeSpan? queuedWriteBudget = null, TimeSpan? writeTimeout = null)
{
var parent = CreateTestProbe();
var actor = parent.ChildActorOf(DriverInstanceActor.Props(
driver, commsLossTimeout: TimeSpan.Zero, maxWritesInFlight: maxWritesInFlight,
queuedWriteBudget: queuedWriteBudget, writeTimeout: writeTimeout));
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
parent.ExpectMsg<DriverInstanceActor.ConnectivityChanged>(m => m.Connected, PresenceBudget);
return (actor, parent);
}
[Fact]
public async Task Two_concurrent_slow_writes_both_complete_within_one_device_round_trip()
{
var driver = new SlowWritableDriver { Latency = WriteLatency };
var (actor, _) = Connect(driver);
var sw = Stopwatch.StartNew();
var askBudget = TimeSpan.FromSeconds(8); // the host's Ask
var first = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-1", 1), askBudget);
var second = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-2", 2), askBudget);
var results = await Task.WhenAll(first, second);
sw.Stop();
results.ShouldAllBe(r => r.Success);
driver.Writes.Select(w => w.FullReference).ShouldBe(new[] { "tag-1", "tag-2" }, ignoreOrder: true);
// Both were inside the driver at once — that is the proof; the wall clock is only a sanity ceiling
// (serialised writes would take two latencies, and the host's Ask is 8 s).
driver.MaxConcurrentWrites.ShouldBe(2, "the second write waited for the first — writes are still serialised on the mailbox");
sw.Elapsed.ShouldBeLessThan(TimeSpan.FromSeconds(8));
}
[Fact]
public async Task Data_changes_are_delivered_while_a_write_is_pending()
{
var driver = new SlowWritableDriver { HoldWrites = true };
var (actor, parent) = Connect(driver);
await actor.Ask<DriverInstanceActor.SubscriptionEstablished>(
new DriverInstanceActor.Subscribe(new[] { "tag-a" }, TimeSpan.FromMilliseconds(100)), PresenceBudget);
var pending = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-a", 7), PresenceBudget);
AwaitCondition(() => driver.WritesEntered == 1, PresenceBudget);
driver.FireDataChange("tag-a", 3.5);
var published = parent.ExpectMsg<DriverInstanceActor.AttributeValueBatchPublished>(PresenceBudget);
published.Items.Single().Value.ShouldBe(3.5);
pending.IsCompleted.ShouldBeFalse("the write is still held — the data change did not wait behind it");
// Positive control: releasing the device completes the write with its real outcome.
driver.ReleaseWrites();
(await pending).Success.ShouldBeTrue();
}
[Fact]
public async Task Writes_beyond_the_in_flight_cap_queue_and_drain_in_order()
{
var driver = new SlowWritableDriver { HoldWrites = true };
var (actor, _) = Connect(driver, maxWritesInFlight: 2);
var asks = Enumerable.Range(0, 4)
.Select(i => actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute($"tag-{i}", i), PresenceBudget))
.ToArray();
AwaitCondition(() => driver.WritesEntered == 2, PresenceBudget);
// Settle: a third write entering the driver would show up here.
await Task.Delay(300);
driver.WritesEntered.ShouldBe(2, "the cap let a third write into the driver");
driver.ReleaseWrites();
var results = await Task.WhenAll(asks);
results.ShouldAllBe(r => r.Success);
// Entry order: the first two slots went to the first two writes, the queued pair entered only after a
// slot freed. Within each pair the order is unspecified — distinct tags run concurrently on the pool;
// only SAME-tag order is guaranteed (pinned by the same-tag test below).
var entered = driver.EnteredRefs;
entered.Count.ShouldBe(4);
entered.Take(2).ShouldBe(new[] { "tag-0", "tag-1" }, ignoreOrder: true);
entered.Skip(2).ShouldBe(new[] { "tag-2", "tag-3" }, ignoreOrder: true);
}
/// <summary>
/// Fix round 1: writes to the SAME tag must keep their order — two concurrent writes v1, v2 could land as
/// v1. A same-tag write waits behind its tag while other tags still run concurrently.
/// </summary>
[Fact]
public async Task Writes_to_the_same_tag_are_serialised_in_order_while_other_tags_run_concurrently()
{
var driver = new SlowWritableDriver { HoldWrites = true };
var (actor, _) = Connect(driver, maxWritesInFlight: 4);
var first = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-a", 1), PresenceBudget);
AwaitCondition(() => driver.WritesEntered == 1, PresenceBudget);
var second = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-a", 2), PresenceBudget);
// Positive control: a DIFFERENT tag is not held behind tag-a.
var other = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-b", 9), PresenceBudget);
AwaitCondition(() => driver.WritesEntered == 2, PresenceBudget);
driver.EnteredRefs.ShouldBe(new[] { "tag-a", "tag-b" });
await Task.Delay(300);
driver.WritesEntered.ShouldBe(2, "the second tag-a write entered the driver while the first was still inside it");
second.IsCompleted.ShouldBeFalse();
driver.ReleaseWrites();
(await Task.WhenAll(first, second, other)).ShouldAllBe(r => r.Success);
driver.Writes.Where(w => w.FullReference == "tag-a").Select(w => w.Value).ShouldBe(new object[] { 1, 2 });
}
/// <summary>
/// Fix round 1: a write that sat in the queue longer than the host's Ask budget has already been reported
/// as a timeout (and reverted) upstream — writing it late would push a stale value to the device.
/// </summary>
[Fact]
public async Task A_queued_write_older_than_the_budget_is_failed_BadTimeout_instead_of_written_late()
{
var driver = new SlowWritableDriver { HoldWrites = true };
var (actor, _) = Connect(driver, maxWritesInFlight: 1, queuedWriteBudget: TimeSpan.FromMilliseconds(300));
var inFlight = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-0", 0), PresenceBudget);
AwaitCondition(() => driver.WritesEntered == 1, PresenceBudget);
var stale = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-1", 1), PresenceBudget);
await Task.Delay(600);
driver.ReleaseWrites();
(await inFlight).Success.ShouldBeTrue();
var staleResult = await stale;
staleResult.Success.ShouldBeFalse();
staleResult.Reason!.ShouldContain("800A0000", customMessage: "BadTimeout");
driver.EnteredRefs.ShouldBe(new[] { "tag-0" }, "the stale write must never reach the device");
}
/// <summary>
/// Fix round 2: a token-ignoring hung write held its tag forever, so every later write to that tag piled up
/// in the queue. The write is bounded pessimistically; on overrun it answers BadTimeout and releases its
/// slot and tag, and a later write to the same tag reaches the device.
/// </summary>
[Fact]
public async Task A_hung_token_ignoring_write_is_abandoned_and_releases_its_tag_for_a_later_write()
{
var driver = new SlowWritableDriver { HoldWrites = true, IgnoreCancellation = true };
var (actor, _) = Connect(driver, maxWritesInFlight: 1, writeTimeout: TimeSpan.FromMilliseconds(300));
var hung = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-a", 1), PresenceBudget);
AwaitCondition(() => driver.WritesEntered == 1, PresenceBudget);
var hungResult = await hung;
hungResult.Success.ShouldBeFalse();
hungResult.Reason!.ShouldContain("800A0000", customMessage: "BadTimeout");
driver.MaxConcurrentWrites.ShouldBe(1);
// The tag (and the single slot) are free again: a later tag-a write reaches the device even though the
// abandoned one is still inside it.
var later = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-a", 2), PresenceBudget);
AwaitCondition(() => driver.WritesEntered == 2, PresenceBudget);
driver.EnteredRefs.ShouldBe(new[] { "tag-a", "tag-a" });
driver.MaxConcurrentWrites.ShouldBe(2, "positive control: the abandoned write really is still wedged inside the driver");
driver.ReleaseWrites();
(await later).Success.ShouldBeTrue();
}
[Fact]
public async Task Queued_writes_fail_fast_when_the_driver_leaves_Connected()
{
var driver = new SlowWritableDriver { HoldWrites = true };
var (actor, parent) = Connect(driver, maxWritesInFlight: 1);
var inFlight = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-0", 0), PresenceBudget);
AwaitCondition(() => driver.WritesEntered == 1, PresenceBudget);
var queued = actor.Ask<DriverInstanceActor.WriteAttributeResult>(
new DriverInstanceActor.WriteAttribute("tag-1", 1), PresenceBudget);
actor.Tell(new DriverInstanceActor.ForceReconnect());
parent.ExpectMsg<DriverInstanceActor.ConnectivityChanged>(m => !m.Connected, PresenceBudget);
var queuedResult = await queued;
queuedResult.Success.ShouldBeFalse();
queuedResult.Reason.ShouldBe("driver not connected");
driver.WritesEntered.ShouldBe(1, "a queued write must never reach a driver that is reconnecting");
// The in-flight one still answers with the device's real outcome once it is released.
driver.ReleaseWrites();
(await inFlight).Success.ShouldBeTrue();
}
/// <summary>A writable + subscribable driver whose writes either take <see cref="Latency"/> or hold until released.</summary>
private sealed class SlowWritableDriver : IDriver, IWritable, ISubscribable
{
private readonly StubHandle _handle = new();
private readonly ConcurrentQueue<WriteRequest> _writes = new();
private readonly ConcurrentQueue<string> _enteredRefs = new();
private TaskCompletionSource _release = new(TaskCreationOptions.RunContinuationsAsynchronously);
private int _entered;
private int _inside;
private int _maxInside;
public TimeSpan Latency { get; init; }
public bool HoldWrites { get; init; }
/// <summary>When true a held write ignores its cancellation token — a wedged driver.</summary>
public bool IgnoreCancellation { get; init; }
public IReadOnlyList<WriteRequest> Writes => _writes.ToArray();
public IReadOnlyList<string> EnteredRefs => _enteredRefs.ToArray();
public int WritesEntered => Volatile.Read(ref _entered);
public int MaxConcurrentWrites => Volatile.Read(ref _maxInside);
public string DriverInstanceId => "slow-writer-1";
public string DriverType => "SlowWriter";
public void ReleaseWrites() => _release.TrySetResult();
public Task InitializeAsync(string driverConfigJson, CancellationToken cancellationToken) => Task.CompletedTask;
public Task ReinitializeAsync(string driverConfigJson, CancellationToken cancellationToken) => Task.CompletedTask;
public Task ShutdownAsync(CancellationToken cancellationToken) => Task.CompletedTask;
public DriverHealth GetHealth() => new(DriverState.Healthy, DateTime.UtcNow, null);
public async Task<IReadOnlyList<WriteResult>> WriteAsync(IReadOnlyList<WriteRequest> writes, CancellationToken cancellationToken)
{
Interlocked.Increment(ref _entered);
foreach (var w in writes) _enteredRefs.Enqueue(w.FullReference);
var inside = Interlocked.Increment(ref _inside);
int seen;
while (inside > (seen = Volatile.Read(ref _maxInside)))
Interlocked.CompareExchange(ref _maxInside, inside, seen);
try
{
if (HoldWrites && IgnoreCancellation) await _release.Task;
else if (HoldWrites) await _release.Task.WaitAsync(cancellationToken);
else if (Latency > TimeSpan.Zero) await Task.Delay(Latency, cancellationToken);
foreach (var w in writes) _writes.Enqueue(w);
return writes.Select(_ => new WriteResult(OpcUaStatus.Good)).ToList();
}
finally
{
Interlocked.Decrement(ref _inside);
}
}
public event EventHandler<DataChangeEventArgs>? OnDataChange;
public Task<ISubscriptionHandle> SubscribeAsync(
IReadOnlyList<string> fullReferences, TimeSpan publishingInterval, CancellationToken cancellationToken) =>
Task.FromResult<ISubscriptionHandle>(_handle);
public Task UnsubscribeAsync(ISubscriptionHandle handle, CancellationToken cancellationToken) => Task.CompletedTask;
public void FireDataChange(string fullRef, object? value) =>
OnDataChange?.Invoke(this, new DataChangeEventArgs(
_handle, fullRef, new DataValueSnapshot(value, OpcUaStatus.Good, DateTime.UtcNow, DateTime.UtcNow)));
}
}
@@ -0,0 +1,173 @@
using System.Diagnostics;
using Akka.Actor;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.OpcUa;
using ZB.MOM.WW.OtOpcUa.Commons.Types;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
/// <summary>
/// Review 02 L1 / P4: a driver child that stops for good must <b>dispose</b> its <see cref="IDriver"/>
/// (the runtime only ever called <c>ShutdownAsync</c>, so a Calculation driver's compiled-script ALCs and an
/// OpcUaClient driver's LoggerFactory + gate + CTS leaked on every respawn), its <c>ShutdownAsync</c> must be
/// bounded (it used to block a dispatcher thread for as long as a dead socket took to answer), and a
/// <b>supervision restart</b> must NOT dispose — the Props closure re-uses the same driver object for the
/// restarted incarnation, which then has to serve through it.
/// </summary>
public sealed class DriverInstanceActorDisposeTests : RuntimeActorTestBase
{
[Fact]
public void Stopping_the_child_shuts_the_driver_down_and_then_disposes_it_exactly_once()
{
var driver = new DisposableRecordingDriver();
var parent = CreateTestProbe();
parent.IgnoreMessages(m => m is DriverInstanceActor.ConnectivityChanged);
var actor = parent.ChildActorOf(DriverInstanceActor.Props(driver));
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
AwaitCondition(() => driver.InitializeCount == 1, PresenceBudget);
Watch(actor);
Sys.Stop(actor);
ExpectTerminated(actor, PresenceBudget);
driver.ShutdownCount.ShouldBe(1);
driver.DisposeCount.ShouldBe(1, "a child stopped for good must dispose the driver it wrapped");
driver.Trace.ToArray().ShouldBe(new[] { "init", "shutdown", "dispose" }, "shutdown must land before dispose");
}
[Fact]
public async Task A_supervision_restart_keeps_the_driver_undisposed_and_the_restarted_child_still_serves()
{
var driver = new DisposableRecordingDriver();
var parent = CreateTestProbe();
// The probe parent's own strategy STOPS a failing child; the production host (no strategy anywhere under
// src/Server) gets Akka's default decider, which RESTARTS it — the route under test.
var actor = parent.ChildActorOf(
DriverInstanceActor.Props(driver), new OneForOneStrategy(_ => Directive.Restart));
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
// Barrier: Subscribe is only handled once Connected (it is dropped in Connecting).
parent.ExpectMsg<DriverInstanceActor.ConnectivityChanged>(m => m.Connected, PresenceBudget);
await actor.Ask<DriverInstanceActor.SubscriptionEstablished>(
new DriverInstanceActor.Subscribe(new[] { "tag-a" }, TimeSpan.FromMilliseconds(100)), PresenceBudget);
// The crash seam (see DriverHostActorRestartTests): a null snapshot NREs on the actor thread, and the
// probe parent's default decider restarts the child in place around the SAME driver object.
driver.FireMalformedDataChange("tag-a");
parent.FishForMessage(m => m is DriverInstanceActor.ReprovisionRequested, PresenceBudget);
driver.ShutdownCount.ShouldBe(1, "PreRestart runs PostStop, which shuts the driver down");
driver.DisposeCount.ShouldBe(0, "a restart re-uses the driver object — disposing it would kill the new incarnation");
// Positive control: the restarted incarnation serves through the undisposed driver. The host would
// re-provision it exactly like this (InitializeRequested + the desired set).
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
parent.FishForMessage(m => m is DriverInstanceActor.ConnectivityChanged { Connected: true }, PresenceBudget);
await actor.Ask<DriverInstanceActor.SubscriptionEstablished>(
new DriverInstanceActor.Subscribe(new[] { "tag-a" }, TimeSpan.FromMilliseconds(100)), PresenceBudget);
driver.FireDataChange("tag-a", 42.0);
var published = (DriverInstanceActor.AttributeValueBatchPublished)parent.FishForMessage(
m => m is DriverInstanceActor.AttributeValueBatchPublished, PresenceBudget);
published.Items.Single().FullReference.ShouldBe("tag-a");
published.Items.Single().Value.ShouldBe(42.0);
// ...and the eventual real stop still disposes it — once.
Watch(actor);
Sys.Stop(actor);
ExpectTerminated(actor, PresenceBudget);
driver.DisposeCount.ShouldBe(1);
}
[Fact]
public void A_hung_ShutdownAsync_is_abandoned_at_the_bound_and_the_driver_is_still_disposed()
{
var never = new TaskCompletionSource(TaskCreationOptions.RunContinuationsAsynchronously);
// Ignores its token on purpose: a wedged socket/COM proxy does not honour cancellation either.
var driver = new DisposableRecordingDriver { ShutdownBehavior = _ => never.Task };
var bound = TimeSpan.FromMilliseconds(300);
var parent = CreateTestProbe();
parent.IgnoreMessages(m => m is DriverInstanceActor.ConnectivityChanged);
var actor = parent.ChildActorOf(DriverInstanceActor.Props(driver, shutdownTimeout: bound));
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
AwaitCondition(() => driver.InitializeCount == 1, PresenceBudget);
Watch(actor);
var sw = Stopwatch.StartNew();
Sys.Stop(actor);
ExpectTerminated(actor, PresenceBudget);
sw.Stop();
driver.ShutdownCount.ShouldBe(1, "the shutdown was attempted (positive control for the bound)");
never.Task.IsCompleted.ShouldBeFalse("the shutdown is still hung — the stop did not wait for it");
sw.Elapsed.ShouldBeLessThan(TimeSpan.FromSeconds(5), "the stop must not be pinned by the hung shutdown");
driver.DisposeCount.ShouldBe(1, "a shutdown that timed out must not skip the dispose");
}
/// <summary>
/// A subscribable driver that records its lifecycle and is <see cref="IAsyncDisposable"/>. Deliberately
/// NOT also <see cref="IDisposable"/>: the actor prefers the async dispose when both exist, and a double
/// implementation would hide a double-dispose.
/// </summary>
private sealed class DisposableRecordingDriver : IDriver, ISubscribable, IAsyncDisposable
{
private readonly StubHandle _handle = new();
private int _initializeCount;
private int _shutdownCount;
private int _disposeCount;
public Func<CancellationToken, Task>? ShutdownBehavior { get; init; }
public List<string> Trace { get; } = new();
public int InitializeCount => Volatile.Read(ref _initializeCount);
public int ShutdownCount => Volatile.Read(ref _shutdownCount);
public int DisposeCount => Volatile.Read(ref _disposeCount);
public string DriverInstanceId => "disposable-1";
public string DriverType => "Disposable";
public Task InitializeAsync(string driverConfigJson, CancellationToken cancellationToken)
{
Interlocked.Increment(ref _initializeCount);
lock (Trace) Trace.Add("init");
return Task.CompletedTask;
}
public Task ReinitializeAsync(string driverConfigJson, CancellationToken cancellationToken) => Task.CompletedTask;
public Task ShutdownAsync(CancellationToken cancellationToken)
{
Interlocked.Increment(ref _shutdownCount);
lock (Trace) Trace.Add("shutdown");
return ShutdownBehavior is { } behavior ? behavior(cancellationToken) : Task.CompletedTask;
}
public DriverHealth GetHealth() => new(DriverState.Healthy, DateTime.UtcNow, null);
public ValueTask DisposeAsync()
{
Interlocked.Increment(ref _disposeCount);
lock (Trace) Trace.Add("dispose");
return ValueTask.CompletedTask;
}
public event EventHandler<DataChangeEventArgs>? OnDataChange;
public Task<ISubscriptionHandle> SubscribeAsync(
IReadOnlyList<string> fullReferences, TimeSpan publishingInterval, CancellationToken cancellationToken) =>
Task.FromResult<ISubscriptionHandle>(_handle);
public Task UnsubscribeAsync(ISubscriptionHandle handle, CancellationToken cancellationToken) => Task.CompletedTask;
public void FireDataChange(string fullRef, object? value) =>
OnDataChange?.Invoke(this, new DataChangeEventArgs(
_handle, fullRef, new DataValueSnapshot(value, OpcUaStatus.Good, DateTime.UtcNow, DateTime.UtcNow)));
/// <summary>The supervision crash seam — see <see cref="RecordingSubscribableDriver.FireMalformedDataChange"/>.</summary>
public void FireMalformedDataChange(string fullRef) =>
OnDataChange?.Invoke(this, new DataChangeEventArgs(_handle, fullRef, null!));
}
}
@@ -0,0 +1,160 @@
using Akka.Actor;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.Types;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
/// <summary>
/// Review 02 H2: the <c>Reconnecting</c> retry timer fired every <c>reconnectInterval</c> whether or not the
/// previous attempt was still running — <c>_initGeneration</c> only discarded the older attempt's RESULT. With an
/// Initialize bound longer than the cadence (Tier C's 15 s vs. the 10 s timer, or the test harness's unbounded
/// invoker) attempt N+1's <c>ShutdownAsync</c> ran while attempt N's <c>InitializeAsync</c> was still inside the
/// driver: the dispose-before-overwrite race the drivers' rule file forbids, once per overlap, for as long as
/// the device stayed down. The retry is now single-flight.
/// </summary>
public sealed class DriverInstanceActorReconnectSingleFlightTests : RuntimeActorTestBase
{
[Fact]
public void Shutdown_and_Initialize_never_overlap_while_the_init_outlasts_the_retry_cadence()
{
// A 400 ms failing init against a 50 ms retry cadence: without the gate the timer starts a new
// Shutdown+Initialize pair every 50 ms on top of the one still running (eight-deep overlap).
var driver = new InterleavingDriver(initDuration: TimeSpan.FromMilliseconds(400), failUntilAttempt: 4);
var parent = CreateTestProbe();
var actor = parent.ChildActorOf(DriverInstanceActor.Props(
driver, reconnectInterval: TimeSpan.FromMilliseconds(50), commsLossTimeout: TimeSpan.Zero));
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
// Positive control: the gate skips ticks, it does not stop retrying — the 4th attempt succeeds and the
// actor connects.
parent.FishForMessage(m => m is DriverInstanceActor.ConnectivityChanged { Connected: true }, PresenceBudget);
driver.Attempts.ShouldBeGreaterThanOrEqualTo(4);
driver.Overlaps.ShouldBe(0,
$"ShutdownAsync entered while an InitializeAsync was still running. Interleaving: {driver.Interleaving}");
}
[Fact]
public void A_pre_reconnect_teardown_that_hangs_is_abandoned_at_the_bound_and_the_init_still_runs()
{
var never = new TaskCompletionSource(TaskCreationOptions.RunContinuationsAsynchronously);
var driver = new InterleavingDriver(initDuration: TimeSpan.Zero, failUntilAttempt: 2)
{
// Only the SECOND shutdown (the reconnect teardown) hangs; the first attempt has none.
ShutdownBehavior = _ => never.Task,
};
var parent = CreateTestProbe();
var actor = parent.ChildActorOf(DriverInstanceActor.Props(
driver, reconnectInterval: TimeSpan.FromMilliseconds(50), commsLossTimeout: TimeSpan.Zero,
shutdownTimeout: TimeSpan.FromMilliseconds(200)));
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
parent.FishForMessage(m => m is DriverInstanceActor.ConnectivityChanged { Connected: true }, PresenceBudget);
never.Task.IsCompleted.ShouldBeFalse("the teardown is still hung — the attempt did not wait for it");
driver.Attempts.ShouldBe(2);
}
/// <summary>
/// Fix round 1: the single-flight gate only cleared on the attempt's RESULT, and the invoker's timeout is
/// token-only — a driver whose InitializeAsync ignores its token wedged the reconnect loop forever. The
/// attempt is now bounded pessimistically (<c>initializeTimeout</c>); an overrun is abandoned, reported
/// as a failure, and the next tick retries.
/// </summary>
[Fact]
public void A_token_ignoring_never_completing_init_is_abandoned_and_a_later_attempt_connects()
{
var driver = new InterleavingDriver(initDuration: TimeSpan.Zero, failUntilAttempt: 1)
{
// Attempt 1 hangs forever and ignores cancellation; attempt 2 completes.
HangOnAttempt = 1,
};
var parent = CreateTestProbe();
var actor = parent.ChildActorOf(DriverInstanceActor.Props(
driver, reconnectInterval: TimeSpan.FromMilliseconds(50), commsLossTimeout: TimeSpan.Zero,
initializeTimeout: TimeSpan.FromMilliseconds(300)));
actor.Tell(new DriverInstanceActor.InitializeRequested("{}"));
parent.FishForMessage(m => m is DriverInstanceActor.ConnectivityChanged { Connected: true }, PresenceBudget);
driver.Attempts.ShouldBe(2);
driver.HungAttemptStillRunning.ShouldBeTrue("positive control: the first attempt really was wedged");
}
/// <summary>Records every Shutdown/Initialize entry and exit so an overlap is a fact, not an inference.</summary>
private sealed class InterleavingDriver : IDriver
{
private readonly TimeSpan _initDuration;
private readonly int _failUntilAttempt;
private readonly Lock _gate = new();
private readonly List<string> _events = new();
private int _attempts;
private int _initsInProgress;
private int _overlaps;
public InterleavingDriver(TimeSpan initDuration, int failUntilAttempt)
{
_initDuration = initDuration;
_failUntilAttempt = failUntilAttempt;
}
public Func<CancellationToken, Task>? ShutdownBehavior { get; init; }
/// <summary>The 1-based attempt that never completes and ignores its token (0 = none).</summary>
public int HangOnAttempt { get; init; }
private readonly TaskCompletionSource _never = new(TaskCreationOptions.RunContinuationsAsynchronously);
public bool HungAttemptStillRunning => !_never.Task.IsCompleted && Volatile.Read(ref _initsInProgress) > 0;
public int Attempts => Volatile.Read(ref _attempts);
public int Overlaps => Volatile.Read(ref _overlaps);
public string Interleaving { get { lock (_gate) return string.Join(" ", _events); } }
public string DriverInstanceId => "interleaving-1";
public string DriverType => "Interleaving";
public async Task InitializeAsync(string driverConfigJson, CancellationToken cancellationToken)
{
var attempt = Interlocked.Increment(ref _attempts);
Interlocked.Increment(ref _initsInProgress);
Record($"init{attempt}>");
try
{
if (attempt == HangOnAttempt) await _never.Task;
if (_initDuration > TimeSpan.Zero) await Task.Delay(_initDuration, CancellationToken.None);
if (attempt < _failUntilAttempt) throw new InvalidOperationException($"init attempt {attempt} refused");
}
finally
{
Record($"<init{attempt}");
Interlocked.Decrement(ref _initsInProgress);
}
}
public Task ReinitializeAsync(string driverConfigJson, CancellationToken cancellationToken) => Task.CompletedTask;
public async Task ShutdownAsync(CancellationToken cancellationToken)
{
if (Volatile.Read(ref _initsInProgress) > 0) Interlocked.Increment(ref _overlaps);
Record("shutdown>");
try
{
if (ShutdownBehavior is { } behavior) await behavior(cancellationToken);
}
finally
{
Record("<shutdown");
}
}
public DriverHealth GetHealth() => new(DriverState.Healthy, DateTime.UtcNow, null);
private void Record(string e) { lock (_gate) _events.Add(e); }
}
}
@@ -47,6 +47,8 @@ public static class V3ArtifactSeeder
/// is parsed out of <c>TagConfig.alarm</c> by <c>TagConfigIntent</c>), which the alarm-ref half of the
/// restart suite needs — an alarm-bearing raw tag is excluded from the VALUE subscription set and carried
/// in the alarm one instead, so the two travel through different child code paths.</param>
/// <param name="driverRowIds">Optional <c>DriverInstanceRowId</c> (the DB row Guid) per driver id; absent
/// ids get <see cref="Guid.Empty"/>, which is what an artifact without the property parses to.</param>
/// <param name="taglessDrivers">Extra <c>DriverInstance</c> ids carrying NO devices and NO tags. The
/// only way to express a driver that is a running child yet absent from both the value- and alarm-ref
/// maps — the <c>_children.Keys</c> arm of <c>ApplyDesiredSubscriptions</c>'s union, which must still
@@ -58,14 +60,16 @@ public static class V3ArtifactSeeder
(string Area, string Line, string Equip, string BackingTag, string? Effective)[] unsRefs,
bool enabled = false,
IReadOnlyDictionary<string, string>? tagConfigByTag = null,
string[]? taglessDrivers = null)
string[]? taglessDrivers = null,
IReadOnlyDictionary<string, Guid>? driverRowIds = null)
{
Guid RowId(string id) => driverRowIds is not null && driverRowIds.TryGetValue(id, out var g) ? g : Guid.Empty;
var drivers = rawTags
.Select(t => (t.Driver, t.DriverName))
.Distinct()
.Select(d => new { DriverInstanceId = d.Driver, RawFolderId = "rf-plant", Name = d.DriverName, DriverType = "Modbus", DriverConfig = "{}", ClusterId = "c1", Enabled = enabled })
.Select(d => new { DriverInstanceRowId = RowId(d.Driver), DriverInstanceId = d.Driver, RawFolderId = "rf-plant", Name = d.DriverName, DriverType = "Modbus", DriverConfig = "{}", ClusterId = "c1", Enabled = enabled })
.Concat((taglessDrivers ?? Array.Empty<string>())
.Select(id => new { DriverInstanceId = id, RawFolderId = "rf-plant", Name = id, DriverType = "Modbus", DriverConfig = "{}", ClusterId = "c1", Enabled = enabled }))
.Select(id => new { DriverInstanceRowId = RowId(id), DriverInstanceId = id, RawFolderId = "rf-plant", Name = id, DriverType = "Modbus", DriverConfig = "{}", ClusterId = "c1", Enabled = enabled }))
.ToArray();
var devices = rawTags
.Select(t => (t.Driver, t.Device))
@@ -201,7 +205,37 @@ public static class V3ArtifactSeeder
return (actor, mux, coordinator);
}
private static void Dispatch(IActorRef actor, TestProbe coordinator, DeploymentId deploymentId, RevisionHash rev)
/// <summary>The same arrangement as <see cref="SpawnHostAndApply"/>, also returning the coordinator probe so a
/// test can <see cref="Dispatch"/> FURTHER deployments to the same host (remove a driver, re-add it).</summary>
/// <param name="kit">The calling TestKit, supplying the ActorSystem + probes.</param>
/// <param name="db">The factory whose deployment was seeded by <see cref="SeedV3Deployment"/>.</param>
/// <param name="localNode">The host's local node id.</param>
/// <param name="deploymentId">The deployment to dispatch.</param>
/// <param name="rev">The revision hash the deployment was sealed under.</param>
/// <param name="driverFactory">Optional driver factory forwarded to <c>DriverHostActor.Props</c>.</param>
/// <param name="reprovisionBackoff">Optional per-driver supervision-reprovision rate limit.</param>
/// <returns>The host actor plus the publish, mux and coordinator probes wired into it.</returns>
public static (IActorRef Actor, TestProbe Publish, TestProbe Mux, TestProbe Coordinator) SpawnHostAndApplyWithCoordinator(
TestKitBase kit, IDbContextFactory<OtOpcUaConfigDbContext> db,
NodeId localNode, DeploymentId deploymentId, RevisionHash rev,
IDriverFactory? driverFactory = null,
TimeSpan? reprovisionBackoff = null)
{
var publish = kit.CreateCleanOutcomePublishProbe();
var (actor, mux, coordinator) = Spawn(kit, db, localNode, publish.Ref, driverFactory, reprovisionBackoff);
Dispatch(actor, coordinator, deploymentId, rev);
publish.ExpectMsg<OpcUaPublishActor.RebuildAddressSpace>(RuntimeActorTestBase.PresenceBudget);
return (actor, publish, mux, coordinator);
}
/// <summary>Dispatches <paramref name="deploymentId"/> to <paramref name="actor"/> and waits for its Applied ack.</summary>
/// <param name="actor">The host.</param>
/// <param name="coordinator">The coordinator probe the host acks to.</param>
/// <param name="deploymentId">The deployment to dispatch.</param>
/// <param name="rev">Its revision hash.</param>
public static void Dispatch(IActorRef actor, TestProbe coordinator, DeploymentId deploymentId, RevisionHash rev)
{
actor.Tell(new DispatchDeployment(deploymentId, rev, CorrelationId.NewId()));
coordinator.ExpectMsg<ApplyAck>(RuntimeActorTestBase.PresenceBudget)
@@ -0,0 +1,23 @@
using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
/// <summary>
/// Builds the ONE message shape a driver child sends for its values — a
/// <see cref="DriverInstanceActor.AttributeValueBatchPublished"/> — for tests that push a single value into the
/// dependency mux by hand. The per-value <see cref="DriverInstanceActor.AttributeValuePublished"/> has no
/// production sender (02/U4); tests should not keep one alive either.
/// </summary>
public static class ValuePublishes
{
/// <summary>A one-item batch for <paramref name="fullReference"/>.</summary>
/// <param name="driverInstanceId">The publishing driver's id.</param>
/// <param name="fullReference">The value's reference (RawPath).</param>
/// <param name="value">The value.</param>
/// <param name="statusCode">The full OPC UA status code.</param>
/// <param name="timestampUtc">The source timestamp.</param>
/// <returns>The batch message.</returns>
public static DriverInstanceActor.AttributeValueBatchPublished Single(
string driverInstanceId, string fullReference, object? value, uint statusCode, DateTime timestampUtc) =>
new(driverInstanceId, new[] { new DriverInstanceActor.AttributeValueItem(fullReference, value, statusCode, timestampUtc) });
}
@@ -0,0 +1,73 @@
using Akka.Actor;
using Microsoft.EntityFrameworkCore;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Configuration;
using ZB.MOM.WW.OtOpcUa.Runtime.Health;
using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Health;
/// <summary>
/// Review 02 P7: <see cref="DbHealthProbeActor"/> ran its SELECT synchronously on the actor thread, so a
/// hanging SQL connect/command held the dispatcher thread for the implicit 15 s / 30 s SqlClient timeouts and
/// queued every later tick (and every <c>GetStatus</c> Ask) behind it. The probe now runs off the mailbox
/// and pipes its outcome back, so the cached answer stays available while a probe hangs.
/// </summary>
public sealed class DbHealthProbeOffThreadTests : RuntimeActorTestBase
{
[Fact]
public async Task GetStatus_answers_from_cache_while_a_probe_is_hung()
{
var factory = new HangingDbFactory();
try
{
var actor = Sys.ActorOf(DbHealthProbeActor.Props(factory));
AwaitCondition(() => factory.Entered > 0, PresenceBudget);
var status = await actor.Ask<DbHealthProbeActor.DbHealthStatus>(
DbHealthProbeActor.GetStatus.Instance, TimeSpan.FromSeconds(3));
status.Reachable.ShouldBeFalse();
status.LastError.ShouldBe("not probed yet");
factory.Entered.ShouldBe(1, "a second tick must not stack a second probe behind a hung one");
}
finally
{
factory.Release();
}
}
[Fact]
public async Task A_probe_that_completes_updates_the_cached_status()
{
// Positive control for the off-thread path: the piped-back outcome lands in the cache.
var db = NewInMemoryDbFactory();
var actor = Sys.ActorOf(DbHealthProbeActor.Props(db));
await AwaitAssertAsync(async () =>
{
var status = await actor.Ask<DbHealthProbeActor.DbHealthStatus>(
DbHealthProbeActor.GetStatus.Instance, TimeSpan.FromSeconds(3));
status.Reachable.ShouldBeTrue();
}, PresenceBudget);
}
/// <summary>Blocks every <c>CreateDbContext</c> until released — a connect that never answers.</summary>
private sealed class HangingDbFactory : IDbContextFactory<OtOpcUaConfigDbContext>
{
private readonly ManualResetEventSlim _release = new(false);
private int _entered;
public int Entered => Volatile.Read(ref _entered);
public void Release() => _release.Set();
public OtOpcUaConfigDbContext CreateDbContext()
{
Interlocked.Increment(ref _entered);
_release.Wait();
throw new InvalidOperationException("released without a database");
}
}
}
@@ -15,18 +15,22 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Health;
public sealed class HealthProbeActorTests : RuntimeActorTestBase
{
/// <summary>Verifies that the DB health probe actor returns reachable status against an in-memory database.</summary>
/// <summary>Verifies that the DB health probe actor returns reachable status against an in-memory database.
/// The probe runs off the mailbox (02/P7), so the first Ask may still see "not probed yet" — poll.</summary>
[Fact]
public async Task DbHealthProbeActor_returns_reachable_against_in_memory_db()
{
var db = NewInMemoryDbFactory();
var actor = Sys.ActorOf(DbHealthProbeActor.Props(db));
var status = await actor.Ask<DbHealthProbeActor.DbHealthStatus>(
DbHealthProbeActor.GetStatus.Instance, PresenceBudget);
await AwaitAssertAsync(async () =>
{
var status = await actor.Ask<DbHealthProbeActor.DbHealthStatus>(
DbHealthProbeActor.GetStatus.Instance, PresenceBudget);
status.Reachable.ShouldBeTrue();
status.LastError.ShouldBeNull();
status.Reachable.ShouldBeTrue();
status.LastError.ShouldBeNull();
}, PresenceBudget);
}
/// <summary>Verifies that the peer OPC UA probe actor reports Ok true against a live listener.</summary>
@@ -0,0 +1,69 @@
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.Messages.Drivers;
using ZB.MOM.WW.OtOpcUa.Runtime.Telemetry;
namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Telemetry;
/// <summary>
/// Review 02 L4: <see cref="TelemetryLocalHub.Evict"/>'s tombstone set only ever shrank on a re-spawn of the
/// SAME id, so a driver removed for good kept its id forever. The tombstone is load-bearing (#609 — it holds
/// back the removed child's late writes), so it cannot simply be cleared; it is now timestamped and swept
/// once older than <see cref="TelemetryLocalHub.EvictionTombstoneTtl"/>, which outlives any in-flight
/// capability call by a wide margin.
/// </summary>
public sealed class TelemetryLocalHubTombstoneSweepTests
{
private static DriverHealthChanged Health(string instanceId) =>
new("cluster-a", instanceId, "Healthy", null, null, 0, DateTime.UtcNow);
private static List<TelemetryItem> Drain(ITelemetrySubscription sub)
{
var items = new List<TelemetryItem>();
while (sub.Reader.TryRead(out var item)) items.Add(item);
return items;
}
[Fact]
public void A_tombstone_holds_late_writes_until_the_ttl_then_is_swept()
{
var clock = new ManualTime { Now = new DateTimeOffset(2026, 10, 22, 12, 0, 0, TimeSpan.Zero) };
var hub = new TelemetryLocalHub(clock);
hub.Emit(new TelemetryItem.Health(Health("drv-a")));
hub.Evict("drv-a");
// Inside the TTL: the late write is dropped (the #609 contract) — and another eviction's sweep does
// not lift a tombstone that is still young.
clock.Now += TelemetryLocalHub.EvictionTombstoneTtl / 2;
hub.Evict("drv-b");
hub.Emit(new TelemetryItem.Health(Health("drv-a")));
using (var sub = hub.Subscribe(16))
Drain(sub).ShouldBeEmpty("a late write inside the TTL must still be held back");
// Past the TTL: the next eviction sweeps drv-a's tombstone and its writes land again.
clock.Now += TelemetryLocalHub.EvictionTombstoneTtl;
hub.Evict("drv-c");
hub.Emit(new TelemetryItem.Health(Health("drv-a")));
using (var sub = hub.Subscribe(16))
Drain(sub).OfType<TelemetryItem.Health>().Single().E.DriverInstanceId.ShouldBe("drv-a");
}
[Fact]
public void Readmit_lifts_a_tombstone_before_its_ttl()
{
var clock = new ManualTime { Now = DateTimeOffset.UtcNow };
var hub = new TelemetryLocalHub(clock);
hub.Evict("drv-a");
hub.Readmit("drv-a");
hub.Emit(new TelemetryItem.Health(Health("drv-a")));
using var sub = hub.Subscribe(16);
Drain(sub).Count.ShouldBe(1);
}
private sealed class ManualTime : TimeProvider
{
public DateTimeOffset Now { get; set; }
public override DateTimeOffset GetUtcNow() => Now;
}
}