ClawdDotNet/tests/ClawdDotNet.Web.Tests/ShutdownTests.cs

using System.Diagnostics;
using ClawdDotNet.App;
using ClawdDotNet.App.Services;
using ClawdDotNet.App.Settings;
using ClawdDotNet.Core.Config;
using ClawdDotNet.Core.Instances;
using ClawdDotNet.Core.Tools;
using Microsoft.Extensions.Logging.Abstractions;
using Shouldly;

namespace ClawdDotNet.Web.Tests;

/// <summary>
/// The web UI must be able to shut down quickly — even with open event streams.
///
/// <para><b>The occasion.</b> An event stream never ends by itself. During an orderly
/// shutdown the browser does not hang up, so <c>RequestAborted</c> does not fire,
/// and Kestrel waits patiently for the request to finish. Only after the deadline
/// of <c>StopAsync</c> was it cut off: five seconds in which the application stood still.
/// It became visible when switching instances — between "switch" and the restart there was this
/// pause every time, as soon as a browser window was open anywhere.</para>
///
/// <para>The streams now additionally hang on <c>ApplicationStopping</c>. This test
/// records the promise so it does not get silently lost in the next rework.</para>
/// </summary>
public sealed class ShutdownTests : IDisposable
{
    private readonly string _root = Path.Combine(
        Path.GetTempPath(), "clawd-tests", "web-" + Guid.NewGuid().ToString("N")[..8]);

    private const string Token = "ein-hinreichend-langes-testtoken";

    /// <summary>
    /// Generously sized: measured against the deadline of StopAsync (five
    /// seconds). If the test stays below two, the promise is kept without a
    /// slow build server turning it red.
    /// </summary>
    private static readonly TimeSpan Limit = TimeSpan.FromSeconds(2);

    private AppHost NewHost(int port)
    {
        var instancesDir = Path.Combine(_root, "Instances");
        var instancePath = Path.Combine(instancesDir, "Test");
        Directory.CreateDirectory(instancePath);

        return new AppHost
        {
            Settings = new SettingsManager(_root),
            Directories = new InstanceDirectoryManager(instancesDir),
            Instance = new InstanceConfig
            {
                InstanceName = "Test",
                WebUiEnabled = true,
                WebUiBinding = "loopback",
                WebUiToken = Token,
                WebServerPort = port
            },
            InstancePath = instancePath,
            LogDirectory = Path.Combine(_root, "Logs"),
            LoggerFactory = NullLoggerFactory.Instance,
            Tools = new ToolRegistry()
        };
    }

    /// <summary>
    /// A free port from the operating system. A fixed number would let two test runs
    /// fail side by side — and those are exactly the errors you search for the longest.
    /// </summary>
    private static int FreePort()
    {
        using var probe = new System.Net.Sockets.TcpListener(System.Net.IPAddress.Loopback, 0);
        probe.Start();
        var port = ((System.Net.IPEndPoint)probe.LocalEndpoint).Port;
        probe.Stop();
        return port;
    }

    [Fact]
    public async Task Shutdown_does_not_wait_for_an_open_event_stream()
    {
        var port = FreePort();
        await using var host = NewHost(port);

        var web = InstanceHttpHost.TryStart(host, NullLoggerFactory.Instance);
        web.ShouldNotBeNull();

        using var client = new HttpClient { Timeout = TimeSpan.FromSeconds(30) };
        client.DefaultRequestHeaders.Add("Authorization", $"Bearer {Token}");

        // Really open the stream and wait for the first line: only then does
        // Kestrel hold the request, and only then does the test check what it wants to check.
        using var answer = await client.GetAsync(
            $"http://127.0.0.1:{port}/events?token={Token}",
            HttpCompletionOption.ResponseHeadersRead);

        answer.EnsureSuccessStatusCode();

        await using var stream = await answer.Content.ReadAsStreamAsync();
        using var reader = new StreamReader(stream);
        (await reader.ReadLineAsync()).ShouldBe(": bereit");

        var clock = Stopwatch.StartNew();
        await web!.DisposeAsync();
        clock.Stop();

        clock.Elapsed.ShouldBeLessThan(Limit,
            $"Das Beenden dauerte {clock.Elapsed.TotalSeconds:0.0}s — der Ereignisstrom "
            + "hat das Herunterfahren nicht bemerkt.");
    }

    [Fact]
    public async Task Shutdown_is_quick_even_without_an_open_connection()
    {
        var port = FreePort();
        await using var host = NewHost(port);

        var web = InstanceHttpHost.TryStart(host, NullLoggerFactory.Instance);
        web.ShouldNotBeNull();

        var clock = Stopwatch.StartNew();
        await web!.DisposeAsync();
        clock.Stop();

        clock.Elapsed.ShouldBeLessThan(Limit);
    }

    /// <summary>
    /// A taken port costs the web UI, not the instance.
    ///
    /// <para>Previously <c>TryStart</c> returned <c>null</c> here, and with that the
    /// entire HTTP surface was gone. Since the socket belongs to it, that is the wrong
    /// order of priority: the instance manager will reach the instance through it, and losing it because
    /// of a taken port would mean making the instance unreachable because
    /// a second one is already running — i.e. exactly in the case the manager
    /// exists for.</para>
    /// </summary>
    [Fact]
    public async Task An_occupied_port_only_costs_the_web_UI()
    {
        var port = FreePort();

        using var blockierer = new System.Net.Sockets.TcpListener(System.Net.IPAddress.Loopback, port);
        blockierer.Start();

        await using var host = NewHost(port);

        await using var web = InstanceHttpHost.TryStart(host, NullLoggerFactory.Instance);

        web.ShouldNotBeNull();
        web!.Url.ShouldBeNull("Die Weboberfläche konnte nicht binden — das gehört gesagt.");

        // The socket holds anyway.
        using var client = InstanceEndpoint.CreateClient(host.InstancePath);
        (await client.GetAsync("/api/info")).EnsureSuccessStatusCode();
    }

    public void Dispose()
    {
        try { Directory.Delete(_root, recursive: true); }
        catch { /* A leftover test folder does not turn a run red. */ }
    }
}