lmxopcua/tests/Server/ZB.MOM.WW.OtOpcUa.Host.Tests/OpcUa/OtOpcUaServerHostedServiceStartFailureTests.cs

using System.Net;
using System.Net.Sockets;
using Akka.Hosting;
using Microsoft.AspNetCore.Builder;
using Microsoft.AspNetCore.Hosting;
using Microsoft.Extensions.Configuration;
using Microsoft.Extensions.DependencyInjection;
using Microsoft.Extensions.Diagnostics.HealthChecks;
using Microsoft.Extensions.Logging;
using Microsoft.Extensions.Logging.Abstractions;
using Microsoft.Extensions.Options;
using Shouldly;
using Xunit;
using ZB.MOM.WW.OtOpcUa.Cluster;
using ZB.MOM.WW.OtOpcUa.Commons.OpcUa;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Core.Authorization;
using ZB.MOM.WW.OtOpcUa.Host.Health;
using ZB.MOM.WW.OtOpcUa.Host.OpcUa;
using ZB.MOM.WW.OtOpcUa.OpcUaServer;
using ZB.MOM.WW.OtOpcUa.OpcUaServer.Security;

namespace ZB.MOM.WW.OtOpcUa.Host.Tests.OpcUa;

/// <summary>
///     R03:2.1 / 1.4 — a failed OPC UA SDK start used to be one Error line: the node stayed Ready, joined
///     its mesh, and acked deploys through the Null sink while clients got connection-refused. The start
///     outcome is now a fact (<see cref="IOpcUaEndpointState"/>) the ready/active tiers read, a driver-only
///     node rethrows so the process exits, and <c>StopAsync</c> actually stops the SDK.
/// </summary>
public sealed class OtOpcUaServerHostedServiceStartFailureTests
{
    private static readonly TimeSpan Deadline = TimeSpan.FromSeconds(30);

    /// <summary>
    ///     Port already bound ⇒ the SDK cannot listen ⇒ the state is Failed, the probe is Unhealthy and the
    ///     real <c>/health/ready</c> endpoint answers 503 — while StartAsync itself returns (fused node).
    ///     Positive control: the same endpoint answers 200 once the state is Started.
    /// </summary>
    [Fact]
    public async Task Port_in_use_marks_the_endpoint_failed_and_health_ready_answers_503()
    {
        using var deadline = new CancellationTokenSource(Deadline);
        using var squatter = SquatAllInterfaces(out var port);
        var pki = NewPki();
        var state = new OpcUaEndpointState();
        var fused = EffectiveRoles.Resolve(new[] { "admin", "driver" }, envRoles: null);

        var service = NewService(port, pki, state, fused);
        try
        {
            await service.StartAsync(deadline.Token);   // must NOT throw on a fused node

            state.Status.ShouldBe(OpcUaEndpointStatus.Failed);
            state.FailureReason.ShouldNotBeNullOrWhiteSpace();

            var probe = await new OpcUaEndpointHealthCheck(state).CheckHealthAsync(new HealthCheckContext());
            probe.Status.ShouldBe(HealthStatus.Unhealthy);
            probe.Description.ShouldNotBeNull();
            probe.Description.ShouldContain("failed to start");

            await using var web = BuildHealthHost(state);
            await web.StartAsync(deadline.Token);
            using var http = new HttpClient { BaseAddress = new Uri(web.Urls.First()) };

            (await http.GetAsync("/health/ready", deadline.Token)).StatusCode.ShouldBe(HttpStatusCode.ServiceUnavailable);
            (await http.GetAsync("/health/active", deadline.Token)).StatusCode.ShouldBe(HttpStatusCode.ServiceUnavailable);

            // Positive control on the endpoint: the same host answers 200 for a started endpoint, so the
            // 503 above measured the probe and not a broken health pipeline.
            state.MarkStarted();
            (await http.GetAsync("/health/ready", deadline.Token)).StatusCode.ShouldBe(HttpStatusCode.OK);
            await web.StopAsync(deadline.Token);
        }
        finally
        {
            await service.DisposeAsync();
            squatter.Stop();
            Cleanup(pki);
        }
    }

    /// <summary>A driver-only node has nothing else to stay up for: the failed start propagates.</summary>
    [Fact]
    public async Task Driver_only_node_rethrows_when_the_sdk_fails_to_start()
    {
        using var deadline = new CancellationTokenSource(Deadline);
        using var squatter = SquatAllInterfaces(out var port);
        var pki = NewPki();
        var state = new OpcUaEndpointState();
        var driverOnly = EffectiveRoles.Resolve(new[] { "driver" }, envRoles: null);

        var service = NewService(port, pki, state, driverOnly);
        try
        {
            await Should.ThrowAsync<Exception>(() => service.StartAsync(deadline.Token));
            state.Status.ShouldBe(OpcUaEndpointStatus.Failed);
        }
        finally
        {
            await service.DisposeAsync();
            squatter.Stop();
            Cleanup(pki);
        }
    }

    /// <summary>
    ///     A clean start is Started and listening; <c>StopAsync</c> closes the port (connection refused, not
    ///     a stale-but-listening endpoint) and marks Stopped; <c>DisposeAsync</c> twice afterwards is a no-op.
    /// </summary>
    [Fact]
    public async Task StopAsync_closes_the_port_and_DisposeAsync_is_idempotent()
    {
        using var deadline = new CancellationTokenSource(Deadline);
        var port = FreePort();
        var pki = NewPki();
        var state = new OpcUaEndpointState();
        var service = NewService(port, pki, state, EffectiveRoles.Resolve(new[] { "driver" }, envRoles: null));
        try
        {
            await service.StartAsync(deadline.Token);
            state.Status.ShouldBe(OpcUaEndpointStatus.Started);
            using (var client = new TcpClient())
            {
                await client.ConnectAsync(IPAddress.Loopback, port, deadline.Token);
                client.Connected.ShouldBeTrue("positive control: the endpoint is listening after StartAsync");
            }

            await service.StopAsync(deadline.Token);

            state.Status.ShouldBe(OpcUaEndpointStatus.Stopped);
            using (var client = new TcpClient())
            {
                await Should.ThrowAsync<SocketException>(async () => await client.ConnectAsync(IPAddress.Loopback, port, deadline.Token));
            }

            await service.DisposeAsync();
            await service.DisposeAsync();   // idempotent
        }
        finally
        {
            await service.DisposeAsync();
            Cleanup(pki);
        }
    }

    private static OtOpcUaServerHostedService NewService(int port, string pki, OpcUaEndpointState state, EffectiveRoles roles) =>
        new(
            Options.Create(new OpcUaApplicationHostOptions
            {
                ApplicationName = "OtOpcUa.StartFailure",
                ApplicationUri = $"urn:OtOpcUa.StartFailure:{Guid.NewGuid():N}",
                OpcUaPort = port, PublicHostname = "localhost", PkiStoreRoot = pki,
            }),
            new DeferredAddressSpaceSink(), new DeferredServiceLevelPublisher(),
            NullOpcUaUserAuthenticator.Instance, NullHistorianDataSource.Instance,
            EmptyDataPlaneAuthoritySource.Instance, new ConfigurationBuilder().Build(),
            () => throw new NotSupportedException("no actor system required"), new ActorRegistry(),
            null, Options.Create(new AclOptions()), TimeProvider.System, NullLoggerFactory.Instance,
            endpointState: state, roles: roles);

    /// <summary>The production health registration (driver role) over the supplied state, on a loopback Kestrel.</summary>
    private static WebApplication BuildHealthHost(OpcUaEndpointState state)
    {
        var builder = WebApplication.CreateBuilder();
        builder.WebHost.UseUrls("http://127.0.0.1:0");
        builder.Logging.ClearProviders();
        builder.Services.AddSingleton<IOpcUaEndpointState>(state);
        builder.Services.AddOtOpcUaHealth(hasAdmin: false, hasDriver: true);
        // The akka + cluster-primary probes need a running ActorSystem; keep only the probe under test so
        // the status measured is the endpoint's. (The registration shape itself is pinned separately.)
        builder.Services.Configure<HealthCheckServiceOptions>(o =>
        {
            foreach (var r in o.Registrations.Where(r => r.Name != OpcUaEndpointHealthCheck.Name).ToList())
                o.Registrations.Remove(r);
        });
        var app = builder.Build();
        app.MapOtOpcUaHealth();
        return app;
    }

    /// <summary>
    ///     Holds a port on the IPv4 AND IPv6 wildcard addresses. The SDK binds its listener on the
    ///     wildcards, and on macOS a loopback-only squatter does not conflict with a wildcard bind (both
    ///     sockets carry SO_REUSEADDR), so a loopback squatter would let the SDK start — the first run of
    ///     this test proved exactly that. The SDK also only fails when NEITHER family binds.
    /// </summary>
    private static TcpListener SquatAllInterfaces(out int port)
    {
        var squatter = new TcpListener(IPAddress.IPv6Any, 0);
        squatter.Server.DualMode = true;
        squatter.ExclusiveAddressUse = true;
        squatter.Start();
        port = ((IPEndPoint)squatter.LocalEndpoint).Port;
        return squatter;
    }

    private static int FreePort()
    {
        using var probe = new TcpListener(IPAddress.Loopback, 0);
        probe.Start();
        var port = ((IPEndPoint)probe.LocalEndpoint).Port;
        probe.Stop();
        return port;
    }

    private static string NewPki() => Path.Combine(Path.GetTempPath(), $"otopcua-startfail-{Guid.NewGuid():N}");

    private static void Cleanup(string pki)
    {
        if (Directory.Exists(pki)) Directory.Delete(pki, recursive: true);
    }
}