lmxopcua/tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.FOCAS.IntegrationTests/Series/WireBackendHealthAndLayoutTests.cs

using System.Net;
using System.Net.Sockets;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Driver.FOCAS.Wire;

namespace ZB.MOM.WW.OtOpcUa.Driver.FOCAS.IntegrationTests.Series;

/// <summary>
///     #740 fix round 1 (review I2) — wire-level cases for H15 (fixed-tree health) and H20 (alarm record layout)
///     against the focas-mock fixture.
/// </summary>
[Collection(FocasSimCollection.Name)]
public sealed class WireBackendHealthAndLayoutTests
{
    private readonly FocasSimFixture _fx;

    /// <summary>Initializes a new instance of the test class with the FOCAS simulation fixture.</summary>
    /// <param name="fx">The FOCAS simulation fixture.</param>
    public WireBackendHealthAndLayoutTests(FocasSimFixture fx) => _fx = fx;

    /// <summary>
    ///     A local TCP relay in front of the shared mock, so ONE test can take "the CNC" away and bring it back
    ///     without stopping a fixture other suites use: <see cref="Down"/> closes every relayed session and stops
    ///     listening (new connects are refused); <see cref="Up"/> listens again on the same port.
    /// </summary>
    private sealed class CncRelay : IAsyncDisposable
    {
        private readonly string _targetHost;
        private readonly int _targetPort;
        private readonly List<TcpClient> _live = [];
        private TcpListener? _listener;
        private CancellationTokenSource? _cts;

        public CncRelay(string targetHost, int targetPort)
        {
            _targetHost = targetHost;
            _targetPort = targetPort;
            Port = Listen(0);
        }

        public int Port { get; private set; }

        private int Listen(int port)
        {
            _cts = new CancellationTokenSource();
            _listener = new TcpListener(IPAddress.Loopback, port);
            _listener.Server.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.ReuseAddress, true);
            _listener.Start();
            var ct = _cts.Token;
            var listener = _listener;
            _ = Task.Run(() => AcceptLoopAsync(listener, ct), CancellationToken.None);
            return ((IPEndPoint)_listener.LocalEndpoint).Port;
        }

        private async Task AcceptLoopAsync(TcpListener listener, CancellationToken ct)
        {
            while (!ct.IsCancellationRequested)
            {
                TcpClient inbound;
                try { inbound = await listener.AcceptTcpClientAsync(ct); }
                catch { return; }
                var outbound = new TcpClient();
                try { await outbound.ConnectAsync(_targetHost, _targetPort, ct); }
                catch { inbound.Dispose(); outbound.Dispose(); continue; }
                lock (_live) { _live.Add(inbound); _live.Add(outbound); }
                _ = Pump(inbound, outbound, ct);
                _ = Pump(outbound, inbound, ct);
            }
        }

        private static async Task Pump(TcpClient from, TcpClient to, CancellationToken ct)
        {
            try { await from.GetStream().CopyToAsync(to.GetStream(), ct); }
            catch { /* the relay was taken down, or a side closed */ }
            finally { from.Dispose(); to.Dispose(); }
        }

        public void Down()
        {
            _cts?.Cancel();
            _listener?.Stop();
            lock (_live)
            {
                foreach (var c in _live) c.Dispose();
                _live.Clear();
            }
        }

        /// <summary>
        ///     Listens again on the SAME port (the driver's host string names it). The re-bind is retried briefly: the
        ///     OS can hold the port for a moment after <see cref="Down"/>, or hand it to another process — the latter
        ///     surfaces as a clear failure after the retries rather than a random flake.
        /// </summary>
        public void Up()
        {
            for (var attempt = 1; ; attempt++)
            {
                try { Listen(Port); return; }
                catch (SocketException) when (attempt < 20) { Thread.Sleep(100); }
            }
        }

        public ValueTask DisposeAsync()
        {
            Down();
            return ValueTask.CompletedTask;
        }
    }

    /// <summary>
    ///     H15 over the wire: a driver serving only <c>FIXED:</c> tags is Healthy with a moving LastSuccessfulRead
    ///     while its CNC answers; when the CNC disappears it degrades with LastSuccessfulRead frozen (the #527
    ///     watchdog's conjunction); and it recovers to Healthy when the CNC returns (positive control).
    /// </summary>
    [Fact]
    public async Task Fixed_only_driver_degrades_on_cnc_loss_and_recovers_over_the_wire()
    {
        if (_fx.SkipReason is not null) Assert.Skip(_fx.SkipReason);
        var ct = TestContext.Current.CancellationToken;
        await _fx.LoadProfileAsync("FWLIB64", ct);

        await using var relay = new CncRelay(_fx.Host, _fx.Port);
        var host = $"focas://127.0.0.1:{relay.Port}";
        var drv = new FocasDriver(new FocasDriverOptions
        {
            Devices = [new FocasDeviceOptions(host)],
            RawTags = FocasRawTags.Entries(
                new FocasTagDefinition("Series", host, "FIXED:Identity/SeriesNumber", FocasDataType.String, Writable: false)),
            Probe = new FocasProbeOptions { Enabled = false },
            Timeout = TimeSpan.FromSeconds(1),
            FixedTree = new FocasFixedTreeOptions
            {
                PollInterval = TimeSpan.FromMilliseconds(50),
                ProgramPollInterval = TimeSpan.Zero,
                TimerPollInterval = TimeSpan.Zero,
            },
        }, driverInstanceId: "wire-fixed-health", clientFactory: new WireFocasClientFactory())
        {
            FixedTreeStaleBudgetOverride = TimeSpan.FromMilliseconds(500),
        };

        await using (drv)
        {
            await drv.InitializeAsync("{}", ct);
            await WaitFor(() => drv.GetDeviceState(host)?.LastFixedPollUtc is not null, TimeSpan.FromSeconds(10));
            var first = drv.GetHealth().LastSuccessfulRead!.Value;
            await WaitFor(() => drv.GetHealth().LastSuccessfulRead > first, TimeSpan.FromSeconds(5));
            drv.GetHealth().State.ShouldBe(DriverState.Healthy);

            relay.Down();
            await WaitFor(() => drv.GetHealth().State == DriverState.Degraded, TimeSpan.FromSeconds(15));
            var frozen = drv.GetHealth().LastSuccessfulRead;
            await Task.Delay(TimeSpan.FromSeconds(1), ct);
            drv.GetHealth().State.ShouldBe(DriverState.Degraded);
            drv.GetHealth().LastSuccessfulRead.ShouldBe(frozen, "nothing answered, so LastSuccessfulRead stays frozen");
            (await drv.ReadAsync(["Series"], ct))[0].StatusCode.ShouldBe(FocasStatusMapper.BadCommunicationError,
                "the stale fixed-tree value reads Bad on the same budget that degrades the driver");

            // Positive control: the CNC returns ⇒ Healthy again, LastSuccessfulRead moves. The connect backoff
            // (1 s doubling) bounds how soon the loop re-dials.
            relay.Up();
            await WaitFor(() => drv.GetHealth().State == DriverState.Healthy, TimeSpan.FromSeconds(40));
            drv.GetHealth().LastSuccessfulRead!.Value.ShouldBeGreaterThan(frozen!.Value);
            (await drv.ReadAsync(["Series"], ct))[0].StatusCode.ShouldBe(FocasStatusMapper.Good);
        }
    }

    /// <summary>
    ///     H20 over the wire: the mock answers <c>cnc_rdalmmsg2</c> in the 80-byte LEGACY layout. Nineteen active
    ///     alarms make a 1,520-byte payload — also exactly 20 × 76 — which the old length test decoded as twenty
    ///     garbage vendor records. Positive control: the unambiguous 3-alarm reply decodes the same alarms.
    /// </summary>
    [Fact]
    public async Task Nineteen_legacy_alarms_decode_as_nineteen_over_the_wire()
    {
        if (_fx.SkipReason is not null) Assert.Skip(_fx.SkipReason);
        var ct = TestContext.Current.CancellationToken;
        await _fx.LoadProfileAsync("FWLIB64", ct);

        var client = new WireFocasClientFactory().Create();
        try
        {
            await client.ConnectAsync(FocasHostAddress.TryParse(_fx.DeviceUri)!, TimeSpan.FromSeconds(2), ct);

            await _fx.PatchStateAsync(new { alarms = Alarms(3) }, ct);
            var three = await client.ReadAlarmsAsync(ct);
            three.Count.ShouldBe(3);
            three[2].AlarmNumber.ShouldBe(402);

            await _fx.PatchStateAsync(new { alarms = Alarms(19) }, ct);
            var nineteen = await client.ReadAlarmsAsync(ct);
            nineteen.Count.ShouldBe(19, "a 1,520-byte legacy reply is 19 records, not 20 vendor ones");
            nineteen.Select(a => a.AlarmNumber).ShouldBe(Enumerable.Range(400, 19));
            nineteen[18].Message.ShouldBe("SERVO ALARM 418");
            nineteen[18].Axis.ShouldBe((short)(1 + (18 % 4)));
        }
        finally
        {
            client.Dispose();
            await _fx.PatchStateAsync(new { alarms = Array.Empty<object>() }, ct);
        }

        static object[] Alarms(int n) => [.. Enumerable.Range(0, n)
            .Select(i => (object)new { alm_no = 400 + i, type = 2 + (i % 3), axis = 1 + (i % 4), msg = $"SERVO ALARM {400 + i}" })];
    }

    private static async Task WaitFor(Func<bool> condition, TimeSpan timeout)
    {
        var deadline = DateTime.UtcNow + timeout;
        while (!condition())
        {
            if (DateTime.UtcNow > deadline) throw new TimeoutException("condition not reached");
            await Task.Delay(25);
        }
    }
}