ClawdDotNet/tests/ClawdDotNet.Manager.Tests/InstanceRegistryTests.cs

using ClawdDotNet.Core.Instances;
using Microsoft.Extensions.Logging.Abstractions;
using Shouldly;

namespace ClawdDotNet.Manager.Tests;

/// <summary>
/// The registry of all instances of a machine (#528, stage 3).
///
/// <para>Two promises stand above everything here, and both follow from E-M6: <b>at startup
/// things are taken over, not started</b>, and <b>on exit nothing is ended.</b>
/// If the manager fails, the instances keep running — it must not be the single point
/// this whole project is built against.</para>
/// </summary>
public sealed class InstanceRegistryTests : IAsyncDisposable
{
    private readonly string _root = Path.Combine(
        Path.GetTempPath(), "clawd-tests", "reg-" + Guid.NewGuid().ToString("N")[..8]);

    private readonly FakeLauncher _launcher = new();
    private readonly FakeChannel _channel = new();
    private readonly List<SupervisionEvent> _events = [];

    private readonly InstanceDirectoryManager _directories;

    private static readonly SupervisionOptions Options = new(
        ProbeInterval: TimeSpan.FromMilliseconds(50),
        StartTimeout: TimeSpan.FromMilliseconds(600),
        SilentChecksBeforeAlarm: 3,
        StopGrace: TimeSpan.FromMilliseconds(300),
        SilentKillGrace: TimeSpan.FromMilliseconds(300),
        MaxDrainWait: TimeSpan.FromSeconds(2));

    public InstanceRegistryTests()
    {
        _directories = new InstanceDirectoryManager(Path.Combine(_root, "Instances"));
        // The socket only answers if a process is really running - exactly as in
        // operation (see FakeChannel.Launcher).
        _channel.Launcher = _launcher;
    }

    private InstanceRegistry Register()
    {
        var register = new InstanceRegistry(
            _directories, _launcher, _channel, NullLoggerFactory.Instance, Options);

        register.Observed += _events.Add;
        return register;
    }

    public async ValueTask DisposeAsync()
    {
        await Task.CompletedTask;
        try { Directory.Delete(_root, recursive: true); }
        catch { /* A leftover test folder does not turn a run red. */ }
    }

    // ─── What is there ───

    [Fact]
    public async Task An_empty_instances_directory_gives_an_empty_registry()
    {
        await using var register = Register();
        await register.RefreshAsync();

        register.Instances.ShouldBeEmpty();
    }

    [Fact]
    public async Task Created_instances_are_in_the_registry()
    {
        _directories.CreateInstance("Alpha");
        _directories.CreateInstance("Beta");

        await using var register = Register();
        await register.RefreshAsync();

        register.Instances.Select(i => i.Name).ShouldBe(["Alpha", "Beta"]);
        register.Instances.ShouldAllBe(i => i.Condition == InstanceCondition.Stopped);
    }

    /// <summary>
    /// A manager that starts everything it finds when coming up would, after a
    /// restart of the machine, also start what someone stopped on purpose.
    /// </summary>
    [Fact]
    public async Task Checking_starts_nothing_by_itself()
    {
        _directories.CreateInstance("Alpha");

        await using var register = Register();
        await register.RefreshAsync();

        _launcher.Starts.ShouldBe(0);
        _events.ShouldBeEmpty();
    }

    [Fact]
    public async Task A_cleared_instance_drops_out_of_the_registry()
    {
        var path = _directories.CreateInstance("Vergaenglich");

        await using var register = Register();
        await register.RefreshAsync();
        register.Instances.Count.ShouldBe(1);

        Directory.Delete(path, recursive: true);
        await register.RefreshAsync();

        register.Instances.ShouldBeEmpty();
    }

    // ─── Create, start, stop ───

    [Fact]
    public async Task A_newly_created_instance_is_in_the_registry_at_once_and_not_running()
    {
        await using var register = Register();

        var path = register.Create("Frisch");

        Directory.Exists(path).ShouldBeTrue();
        File.Exists(Path.Combine(path, "InstanceSettings.json")).ShouldBeTrue();

        var instance = register.Instances.ShouldHaveSingleItem();
        instance.Condition.ShouldBe(InstanceCondition.Unknown);
        _launcher.Starts.ShouldBe(0);
    }

    [Fact]
    public async Task Starting_and_stopping_go_through_the_registry()
    {
        _channel.Answer = new InstanceIdentity("i-1", "Alpha", "0.1.2");

        await using var register = Register();
        var path = register.Create("Alpha");

        (await register.StartAsync(path)).ShouldBeTrue();
        register.Instances.Single().Condition.ShouldBe(InstanceCondition.Running);

        // The instance ends upon the request.
        var process = _launcher.Last!;
        _channel.OnStopRequest = _ => { _channel.Answer = null; process.Exit(0); };

        (await register.StopAsync(path)).ShouldBeTrue();
        register.Instances.Single().Condition.ShouldBe(InstanceCondition.Stopped);
    }

    [Fact]
    public async Task An_unknown_directory_is_not_created_silently()
    {
        await using var register = Register();

        await Should.ThrowAsync<KeyNotFoundException>(
            () => register.StartAsync(Path.Combine(_root, "gibtsnicht")));
    }

    // ─── E-M6 ───

    [Fact]
    public async Task Running_instances_are_adopted_when_checking()
    {
        string path;

        // The first manager creates two and starts only one.
        await using (var firstOne = Register())
        {
            path = firstOne.Create("Alpha");
            firstOne.Create("Beta");

            // Only Alpha answers. Information that names the same instance for every
            // directory would turn green here and in operation mix up two
            // instances.
            _channel.AnswerFor[Path.GetFullPath(path)] =
                new InstanceIdentity("i-1", "Alpha", "0.1.2");

            (await firstOne.StartAsync(path)).ShouldBeTrue();
        }

        _events.Clear();

        // Second manager, the same machine.
        await using var secondOne = Register();
        await secondOne.RefreshAsync();

        var alpha = secondOne.Instances.Single(i => i.Name == "Alpha");
        alpha.Condition.ShouldBe(InstanceCondition.Running);
        alpha.ProcessId.ShouldBe(_launcher.Last!.Id);

        // Beta does not answer and stays off - it is not pulled up along with it.
        secondOne.Instances.Single(i => i.Name == "Beta").Condition
            .ShouldBe(InstanceCondition.Stopped);

        // Taking over is not starting.
        _launcher.Starts.ShouldBe(1);
        _events.ShouldBeEmpty();
    }

    /// <summary>
    /// Whoever closes the manager wants to close the manager — not its instances.
    /// </summary>
    [Fact]
    public async Task Stopping_the_manager_stops_no_instance()
    {
        _channel.Answer = new InstanceIdentity("i-1", "Alpha", "0.1.2");

        var register = Register();
        var path = register.Create("Alpha");
        await register.StartAsync(path);

        await register.DisposeAsync();

        _channel.StopRequests.ShouldBe(0);
        _launcher.Last!.WasKilled.ShouldBeFalse();
        _launcher.Last.HasExited.ShouldBeFalse();

        // And the state file stays behind - it is the hint for the next one.
        InstanceStateStore.Read(path).ShouldNotBeNull();
    }

    /// <summary>
    /// The one exception: a license revocation is not an outage, but a verdict.
    /// </summary>
    [Fact]
    public async Task With_a_reason_affecting_everyone_all_are_paused()
    {
        _channel.Answer = new InstanceIdentity("i-1", "Alpha", "0.1.2");

        await using var register = Register();
        var a = register.Create("Alpha");
        await register.StartAsync(a);

        var process = _launcher.Last!;
        _channel.OnStopRequest = _ => { _channel.Answer = null; process.Exit(0); };

        await register.StopAllAsync("Lizenz widerrufen");

        register.Instances.Single().Condition.ShouldBe(InstanceCondition.Stopped);
        _channel.StopRequests.ShouldBe(1);
    }

    // ─── Occurrences ───

    [Fact]
    public async Task Incidents_of_single_instances_reach_the_registry()
    {
        _channel.Answer = new InstanceIdentity("i-1", "Alpha", "0.1.2");

        await using var register = Register();
        var path = register.Create("Alpha");
        await register.StartAsync(path);

        _events.Select(e => e.Kind).ShouldBe([InstanceEventKind.Started]);
        _events[0].InstanceDirectory.ShouldBe(Path.GetFullPath(path));
    }
}