lmxopcua/tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.Modbus.IntegrationTests/CommsRecoveryTests.cs

using System.Diagnostics;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;

namespace ZB.MOM.WW.OtOpcUa.Driver.Modbus.IntegrationTests;

/// <summary>
///     Comms-loss → recovery against the real pymodbus <c>standard</c> fixture, with a loopback
///     <see cref="TcpRelay"/> in front of it so the device can disappear and come back without touching the
///     shared container. Covers the review 07 High findings P3 (transport self-heal) and H2 (deadband
///     recovery publish).
/// </summary>
[Collection(ModbusSimulatorCollection.Name)]
[Trait("Category", "Integration")]
[Trait("Device", "Standard")]
public sealed class CommsRecoveryTests(ModbusSimulatorFixture sim)
{
    private const uint StatusGood = 0u;

    private void SkipUnlessStandardLive()
    {
        if (sim.SkipReason is not null) Assert.Skip(sim.SkipReason);
        var profile = Environment.GetEnvironmentVariable("MODBUS_SIM_PROFILE");
        if (profile is not null && !string.Equals(profile, "standard", StringComparison.OrdinalIgnoreCase))
            Assert.Skip($"MODBUS_SIM_PROFILE={profile} — these tests need the standard profile (HR[0..31] = address).");
    }

    private ModbusDriverOptions Options(TcpRelay relay, params ModbusTagDefinition[] tags) => new()
    {
        Host = "127.0.0.1",
        Port = relay.Port,
        UnitId = 1,
        Timeout = TimeSpan.FromSeconds(2),
        RawTags = ModbusRawTags.Entries(tags),
        Probe = new ModbusProbeOptions { Enabled = false },
    };

    private static async Task<T> WaitFor<T>(Func<Task<T>> probe, Func<T, bool> done, TimeSpan budget)
    {
        var sw = Stopwatch.StartNew();
        T last;
        do
        {
            last = await probe();
            if (done(last)) return last;
            await Task.Delay(200, TestContext.Current.CancellationToken);
        } while (sw.Elapsed < budget);
        return last;
    }

    /// <summary>
    ///     P3: after the device drops (two failed calls in a row, socket torn down, reconnect refused) the
    ///     transport must reconnect ON ITS OWN once the device is back — before this fix every later call
    ///     failed fast with "Transport not connected" until the 90 s comms-loss watchdog re-initialised the
    ///     driver.
    /// </summary>
    [Fact]
    public async Task Driver_reads_again_after_the_device_drops_and_returns_without_reinit()
    {
        SkipUnlessStandardLive();
        await using var relay = new TcpRelay(sim.Host, sim.Port);
        var tag = new ModbusTagDefinition("HR5", ModbusRegion.HoldingRegisters, 5, ModbusDataType.UInt16, Writable: false);
        await using var driver = new ModbusDriver(Options(relay, tag), "modbus-p3");
        await driver.InitializeAsync("{}", TestContext.Current.CancellationToken);

        var good = await driver.ReadAsync(["HR5"], TestContext.Current.CancellationToken);
        good[0].StatusCode.ShouldBe(StatusGood);
        good[0].Value.ShouldBe((ushort)5);

        relay.Cut();
        for (var i = 0; i < 2; i++)
        {
            var bad = await driver.ReadAsync(["HR5"], TestContext.Current.CancellationToken);
            bad[0].StatusCode.ShouldNotBe(StatusGood);
        }
        driver.GetHealth().State.ShouldBe(DriverState.Degraded);

        relay.Restore();
        var recovered = await WaitFor(
            () => driver.ReadAsync(["HR5"], TestContext.Current.CancellationToken),
            r => r[0].StatusCode == StatusGood,
            TimeSpan.FromSeconds(20));
        recovered[0].StatusCode.ShouldBe(StatusGood, "the transport must reconnect by itself once the device is back");
        recovered[0].Value.ShouldBe((ushort)5);
        driver.GetHealth().State.ShouldBe(DriverState.Healthy);
    }

    /// <summary>
    ///     U1 (DisableFC23 wired): a multi-register write on the default (FC23 write + read-back) path
    ///     round-trips against the real server — whether it implements FC23 or answers Illegal Function and
    ///     is moved to FC16 — and so does the forced-FC16 path. Uses the standard profile's scratch HR[200..209].
    /// </summary>
    [Theory]
    [InlineData(false, 200)]
    [InlineData(true, 202)]
    public async Task Multi_register_write_round_trips_with_and_without_FC23(bool disableFc23, int address)
    {
        SkipUnlessStandardLive();
        var tag = new ModbusTagDefinition("F", ModbusRegion.HoldingRegisters, (ushort)address, ModbusDataType.Float32);
        var opts = new ModbusDriverOptions
        {
            Host = sim.Host, Port = sim.Port, UnitId = 1, Timeout = TimeSpan.FromSeconds(2), DisableFC23 = disableFc23,
            RawTags = ModbusRawTags.Entries([tag]), Probe = new ModbusProbeOptions { Enabled = false },
        };
        await using var driver = new ModbusDriver(opts, "modbus-u1");
        await driver.InitializeAsync("{}", TestContext.Current.CancellationToken);

        var value = disableFc23 ? 12.25f : 6.5f;
        for (var i = 0; i < 2; i++) // second write exercises the latched path when the server lacks FC23
        {
            var w = await driver.WriteAsync([new WriteRequest("F", value + i)], TestContext.Current.CancellationToken);
            w[0].StatusCode.ShouldBe(StatusGood);
            var r = await driver.ReadAsync(["F"], TestContext.Current.CancellationToken);
            r[0].Value.ShouldBe(value + i);
        }
        driver.GetHealth().State.ShouldBe(DriverState.Healthy);
    }

    /// <summary>
    ///     H2: a deadbanded tag whose device returns holding the SAME value must publish the Good recovery.
    ///     HR[5] is a constant 5 in the standard profile, so the recovery value is identical by construction.
    ///     Positive control: the Bad publish during the outage is observed first.
    /// </summary>
    [Fact]
    public async Task Deadbanded_subscription_publishes_the_Good_recovery_with_an_identical_value()
    {
        SkipUnlessStandardLive();
        await using var relay = new TcpRelay(sim.Host, sim.Port);
        var tag = new ModbusTagDefinition("HR5", ModbusRegion.HoldingRegisters, 5, ModbusDataType.UInt16,
            Writable: false, Deadband: 10.0);
        await using var driver = new ModbusDriver(Options(relay, tag), "modbus-h2");
        await driver.InitializeAsync("{}", TestContext.Current.CancellationToken);

        var published = new System.Collections.Concurrent.ConcurrentQueue<DataValueSnapshot>();
        driver.OnDataChange += (_, e) => published.Enqueue(e.Snapshot);
        var handle = await driver.SubscribeAsync(["HR5"], TimeSpan.FromMilliseconds(200), TestContext.Current.CancellationToken);
        try
        {
            (await WaitFor(() => Task.FromResult(published.ToArray()),
                p => p.Any(s => s.StatusCode == StatusGood), TimeSpan.FromSeconds(10)))
                .ShouldContain(s => s.StatusCode == StatusGood, "initial Good publish");

            relay.Cut();
            (await WaitFor(() => Task.FromResult(published.ToArray()),
                p => p.Any(s => s.StatusCode != StatusGood), TimeSpan.FromSeconds(20)))
                .ShouldContain(s => s.StatusCode != StatusGood, "the outage must publish Bad (positive control)");

            var badCount = published.Count;
            relay.Restore();
            var after = await WaitFor(() => Task.FromResult(published.Skip(badCount).ToArray()),
                p => p.Any(s => s.StatusCode == StatusGood), TimeSpan.FromSeconds(30));
            var recovery = after.FirstOrDefault(s => s.StatusCode == StatusGood);
            recovery.ShouldNotBeNull("the Good recovery with the same value must not be swallowed by the deadband");
            recovery.Value.ShouldBe((ushort)5);
        }
        finally
        {
            await driver.UnsubscribeAsync(handle, TestContext.Current.CancellationToken);
        }
    }
}