lmxopcua/tests/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer.Tests/NodeManagerLockTimeBenchmarkTests.cs

using System.Diagnostics;
using Microsoft.Extensions.Logging.Abstractions;
using Opc.Ua;
using Opc.Ua.Server;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Commons.OpcUa;

namespace ZB.MOM.WW.OtOpcUa.OpcUaServer.Tests;

/// <summary>
///     Review 01 P1 / P2 / P3 — the lock-time budget of the three teardown paths that run under the
///     manager-wide SDK <c>Lock</c> on the pinned publish thread, where every Read / Browse / Publish waits.
///     <list type="bullet">
///         <item>P1: a full rebuild of a 10k-tag device folder was O(k²) per folder because
///             <c>NodeState.RemoveChild</c> is a linear scan + <c>RemoveAt</c> per leaf (~50M compares).</item>
///         <item>P2: removing M equipment from an N-node space scanned every map per equipment (O(M·N)).</item>
///         <item>P3: every surgical condition removal re-walked every condition for the Server-object
///             monitoring bookkeeping (O(C) per removal) while any <c>i=2253</c> item existed.</item>
///     </list>
///     The wall-clock budgets are generous for CI hardware; the before/after numbers are in the commit
///     message. P3 is pinned by the bookkeeping's own visit counter, which does not depend on the machine.
///     lmxopcua#769: the wall-clock legs only ASSERT their budget when <c>OTOPCUA_BENCH=1</c> (opt-in, like
///     <c>OTOPCUA_SCRIPT_PERF</c>); the default run still executes every teardown and asserts the state correctness.
///     Historical note on the budget: no filter excludes <c>Category=Benchmark</c> (CI drops only E2E + LiveIntegration):
///     measured after the fix at 5–6 ms (P1) and 58–64 ms (P2) on a loaded dev machine against the 1 s budget —
///     ≥ 15× headroom, so it does not flake on slow CI. The flip side: the pre-fix code (75–150 ms / 470–680 ms) also
///     fits 1 s at these sizes, so the wall-clock legs guard against gross (order-of-magnitude) regressions only; the
///     machine-independent P3 visit counter is the strict pin.
/// </summary>
[Trait("Category", "Benchmark")]
public sealed class NodeManagerLockTimeBenchmarkTests(ITestOutputHelper output) : IDisposable
{
    private static CancellationToken Ct => TestContext.Current.CancellationToken;
    // lmxopcua#769: 1 s flaked once in a full parallel run (the 50k-node leg). Measured cost is 5-60 ms, so 5 s keeps
    // >= 80x headroom against a CPU-starved CI box while still catching an order-of-magnitude (O(N^2)) regression.
    private static readonly TimeSpan Budget = TimeSpan.FromSeconds(5);

    // lmxopcua#769: even 5 s flaked in a full parallel run on a starved box, so the wall-clock legs are opt-in, like
    // ScriptRunnerPerfTests (OTOPCUA_SCRIPT_PERF=1): OTOPCUA_BENCH=1 enforces the budget. The default run still executes
    // every teardown and asserts the state correctness (counts, SDK handles, re-population) and logs the measured time.
    private static bool EnforceBudget => Environment.GetEnvironmentVariable("OTOPCUA_BENCH") == "1";

    private void AssertWithinBudget(Stopwatch sw, string what)
    {
        output.WriteLine($"BENCH {what}: {sw.ElapsedMilliseconds} ms (budget {Budget.TotalMilliseconds:F0} ms, enforced={EnforceBudget})");
        if (EnforceBudget)
            sw.Elapsed.ShouldBeLessThan(Budget, $"{what} took {sw.ElapsedMilliseconds} ms");
    }

    private readonly string _pkiRoot = Path.Combine(Path.GetTempPath(), $"otopcua-bench-{Guid.NewGuid():N}");

    /// <summary>P1 — a rebuild that discards one device folder holding 10k tags completes within the budget.</summary>
    [Fact]
    public async Task Rebuild_of_a_10k_tag_folder_stays_within_budget()
    {
        var (host, server) = await BootAsync();
        try
        {
            var nm = server.NodeManager!;
            const string device = "Plant/Modbus/dev1";
            nm.EnsureFolder(device, null, "dev1", AddressSpaceRealm.Raw);
            for (var i = 0; i < 10_000; i++)
                nm.EnsureVariable($"{device}/T{i}", device, $"T{i}", "Int32", writable: false, realm: AddressSpaceRealm.Raw);
            nm.VariableCount.ShouldBe(10_000);

            var sw = Stopwatch.StartNew();
            nm.RebuildAddressSpace();
            sw.Stop();
            nm.CompleteAddressSpaceApply();

            nm.VariableCount.ShouldBe(0);
            nm.FolderCount.ShouldBe(0);
            // Positive control: the space is really empty for the SDK too, and can be re-populated.
            nm.GetManagerHandle(new NodeId($"{device}/T0", nm.NamespaceIndexForRealm(AddressSpaceRealm.Raw))).ShouldBeNull();
            nm.EnsureFolder(device, null, "dev1", AddressSpaceRealm.Raw);
            nm.EnsureVariable($"{device}/T0", device, "T0", "Int32", writable: false, realm: AddressSpaceRealm.Raw);
            nm.TryGetVariable($"{device}/T0", AddressSpaceRealm.Raw)!.Parent.ShouldBeSameAs(nm.TryGetFolder(device, AddressSpaceRealm.Raw));

            AssertWithinBudget(sw, "rebuild 10k-tag folder");
        }
        finally { await host.DisposeAsync(); }
    }

    /// <summary>P2 — removing 200 equipment (250 signals each, 50k nodes) completes within the budget.</summary>
    [Fact]
    public async Task Removing_200_equipment_from_a_50k_node_space_stays_within_budget()
    {
        var (host, server) = await BootAsync();
        try
        {
            var nm = server.NodeManager!;
            const int equipmentCount = 200, perEquipment = 250;
            for (var e = 0; e < equipmentCount; e++)
            {
                var equip = $"area/line/eq{e}";
                nm.EnsureFolder(equip, null, $"eq{e}", AddressSpaceRealm.Uns);
                for (var i = 0; i < perEquipment; i++)
                    nm.EnsureVariable($"{equip}/S{i}", equip, $"S{i}", "Float", writable: false, realm: AddressSpaceRealm.Uns);
            }
            // A survivor the removals must not touch.
            nm.EnsureFolder("area/line/keep", null, "keep", AddressSpaceRealm.Uns);
            nm.EnsureVariable("area/line/keep/S0", "area/line/keep", "S0", "Float", writable: false, realm: AddressSpaceRealm.Uns);
            nm.VariableCount.ShouldBe(equipmentCount * perEquipment + 1);

            var sw = Stopwatch.StartNew();
            for (var e = 0; e < equipmentCount; e++)
                nm.RemoveEquipmentSubtree($"area/line/eq{e}", AddressSpaceRealm.Uns).ShouldBeTrue();
            sw.Stop();

            nm.VariableCount.ShouldBe(1);
            nm.FolderCount.ShouldBe(1);
            nm.TryGetVariable("area/line/keep/S0").ShouldNotBeNull();
            nm.GetManagerHandle(new NodeId("area/line/eq0/S0", nm.NamespaceIndexForRealm(AddressSpaceRealm.Uns))).ShouldBeNull();

            AssertWithinBudget(sw, "remove 200 equipment / 50k nodes");
        }
        finally { await host.DisposeAsync(); }
    }

    /// <summary>
    ///     P3 — with a Server-object item live, a surgical condition removal lowers only that condition's
    ///     monitoring unit instead of re-walking every condition: the bookkeeping visits at most one
    ///     condition per removal. (A full reconcile visits every condition twice — desired + held.)
    /// </summary>
    [Fact]
    public async Task A_surgical_condition_removal_visits_only_its_own_condition()
    {
        var (host, server) = await BootAsync();
        try
        {
            var nm = server.NodeManager!;
            nm.EnsureFolder("eq-1", null, "Equipment 1", AddressSpaceRealm.Uns);
            const int conditions = 50;
            for (var i = 0; i < conditions; i++)
                nm.MaterialiseAlarmCondition($"alm-{i}", "eq-1", $"Alarm {i}", "OffNormalAlarm", 500, AddressSpaceRealm.Uns);

            // A Server-object (all-events) subscriber, so the per-condition monitoring unit is in play.
            nm.SubscribeToAllEvents(
                new OperationContext(new RequestHeader(), secureChannelContext: null, RequestType.CreateMonitoredItems, requestLifetime: RequestLifetime.None, identity: null),
                1, new StubEventItem(), unsubscribe: false);
            nm.ConditionEventsMonitored("alm-0", AddressSpaceRealm.Uns).ShouldBeTrue("non-vacuity: the units really were raised");

            var before = nm.AllEventsMonitoringConditionVisits;
            for (var i = 0; i < 10; i++)
                nm.RemoveAlarmConditionNode($"alm-{i}", AddressSpaceRealm.Uns).ShouldBeTrue();
            var visits = nm.AllEventsMonitoringConditionVisits - before;
            output.WriteLine($"BENCH all-events visits for 10 condition removals over {conditions} conditions: {visits}");

            visits.ShouldBeLessThanOrEqualTo(10, $"ten removals visited {visits} conditions; a full reconcile per removal would visit ~{10 * 2 * conditions}");
            // The survivors keep their unit.
            nm.ConditionEventsMonitored("alm-49", AddressSpaceRealm.Uns).ShouldBeTrue();
        }
        finally { await host.DisposeAsync(); }
    }

    private async Task<(OpcUaApplicationHost Host, OtOpcUaSdkServer Server)> BootAsync() =>
        await OpcUaHostStart.StartAsync(
            port => new OpcUaApplicationHost(
                new OpcUaApplicationHostOptions
                {
                    ApplicationName = "OtOpcUa.BenchTest",
                    ApplicationUri = $"urn:OtOpcUa.BenchTest:{Guid.NewGuid():N}",
                    OpcUaPort = port,
                    PublicHostname = "localhost",
                    PkiStoreRoot = _pkiRoot,
                },
                NullLogger<OpcUaApplicationHost>.Instance),
            () => new OtOpcUaSdkServer(), Ct);

    public void Dispose()
    {
        if (Directory.Exists(_pkiRoot)) Directory.Delete(_pkiRoot, recursive: true);
    }

    /// <summary>The minimum of a Server-object event item the all-events bookkeeping touches: identity only.</summary>
    private sealed class StubEventItem : IEventMonitoredItem
    {
        public bool MonitoringAllEvents => true;
        public NodeId NodeId => ObjectIds.Server;
        public IUserIdentity EffectiveIdentity => null!;
        public object ManagerHandle => null!;
        public void QueueEvent(IFilterTarget instance) { }
        public void QueueEvent(IFilterTarget instance, bool bypassFilter) { }
        public void Dispose() { }

        public IAsyncNodeManager NodeManager => throw new NotSupportedException();
        public ISession Session => throw new NotSupportedException();
        public uint Id => 1;
        public uint SubscriptionId => 1;
        public bool IsDurable => false;
        public uint ClientHandle => 1;
        public ISubscription SubscriptionCallback { get => throw new NotSupportedException(); set => throw new NotSupportedException(); }
        public int MonitoredItemType => MonitoredItemTypeMask.Events;
        public bool IsReadyToPublish => false;
        public bool IsReadyToTrigger { get => false; set { } }
        public bool IsResendData => false;
        public MonitoringMode MonitoringMode => MonitoringMode.Reporting;
        public double SamplingInterval => 0;
        public EventFilter EventFilter => null!;
        public void SetupResendDataTrigger() { }
        public ServiceResult GetCreateResult(out MonitoredItemCreateResult result) => throw new NotSupportedException();
        public ServiceResult GetModifyResult(out MonitoredItemModifyResult result) => throw new NotSupportedException();
        public IStoredMonitoredItem ToStorableMonitoredItem() => throw new NotSupportedException();
        public MonitoringMode SetMonitoringMode(MonitoringMode monitoringMode) => throw new NotSupportedException();
        public bool Publish(OperationContext context, Queue<EventFieldList> notifications, uint maxNotificationsPerPublish) => throw new NotSupportedException();
        public ServiceResult ModifyAttributes(DiagnosticsMasks diagnosticsMasks, TimestampsToReturn timestampsToReturn, uint clientHandle, MonitoringFilter originalFilter, MonitoringFilter filterToUse, Opc.Ua.Range range, double samplingInterval, uint queueSize, bool discardOldest) => throw new NotSupportedException();
    }
}