ClawdDotNet/tests/ClawdDotNet.Manager.Tests/ManagerCoreTests.cs

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

namespace ClawdDotNet.Manager.Tests;

/// <summary>Counts for whom a token was obtained.</summary>
internal sealed class FakeTokenSource : IInstanceTokenSource
{
    public List<string> Fuer { get; } = [];

    public bool Erfolg { get; set; } = true;

    public Task<bool> EnsureTokenAsync(string instanceDirectory, CancellationToken ct = default)
    {
        Fuer.Add(Path.GetFullPath(instanceDirectory));
        return Task.FromResult(Erfolg);
    }
}

/// <summary>
/// The instance manager as a whole (#528, stage 3).
///
/// <para><b>The license is checked here without a license server</b>, and that is intentional: the
/// promise is not "the gatekeeper works" — that is the SDK's job —,
/// but "a verdict leads to stopping all instances, an outage does not". Exactly
/// this path is taken here, triggered via
/// <see cref="ManagerCore.HandleLicenseRevokedAsync"/>.</para>
/// </summary>
public sealed class ManagerCoreTests : IAsyncDisposable
{
    private readonly string _root = Path.Combine(
        Path.GetTempPath(), "clawd-tests", "core-" + Guid.NewGuid().ToString("N")[..8]);

    private readonly FakeLauncher _launcher = new();
    private readonly FakeChannel _channel = new();
    private readonly FakeTokenSource _tokens = new();
    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 ManagerCoreTests()
    {
        Directory.CreateDirectory(_root);
        _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 ManagerCore Manager()
    {
        var register = new InstanceRegistry(
            _directories, _launcher, _channel, NullLoggerFactory.Instance, Options);

        return new ManagerCore(
            new ManagerSettingsStore(_root), register, _tokens, NullLoggerFactory.Instance);
    }

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

    /// <summary>
    /// Lets the instance answer from now on — and checks that it was really
    /// started afterwards and not merely taken over.
    ///
    /// <para><b>The difference is not a subtlety.</b> Whoever stores the answer before
    /// the manager comes up says with it "this instance is already running" — the
    /// manager then takes it over (E-M6), the launcher is never touched, and the test
    /// checks something other than it claims. Exactly that happened to me here twice.</para>
    /// </summary>
    private async Task<StartRefusal> StartAndCheck(ManagerCore manager, string path, string name)
    {
        var before = _launcher.Starts;

        _channel.AnswerFor[Path.GetFullPath(path)] = new InstanceIdentity("i-1", name, "0.1.2");

        var result = await manager.StartInstanceAsync(path);

        _launcher.Starts.ShouldBe(before + 1, "Gestartet, nicht uebernommen.");
        return result;
    }

    // ─── Starting up ───

    [Fact]
    public async Task Without_license_check_the_manager_works()
    {
        // No gatekeeper means no lock - otherwise there would be a chicken-and-egg lock
        // before anyone can even enter a key.
        await using var manager = Manager();

        (await manager.StartAsync()).ShouldBeTrue();
        manager.LicenseHolds.ShouldBeTrue();
    }

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

        await using var manager = Manager();
        await manager.StartAsync();

        manager.Instances.Instances.Count.ShouldBe(1);
        _launcher.Starts.ShouldBe(0);
    }

    // ─── Token ───

    [Fact]
    public async Task A_new_instance_gets_its_own_token_right_away()
    {
        // At creation and not at the first start: then the instance file belongs
        // to nobody without doubt.
        await using var manager = Manager();

        var path = await manager.CreateInstanceAsync("Frisch");

        _tokens.Fuer.ShouldBe([Path.GetFullPath(path)]);
    }

    [Fact]
    public async Task Before_the_start_the_token_is_fetched()
    {
        // The case of an instance that existed before the manager - or for which the
        // Deploymentcenter was not reachable at creation.
        var path = _directories.CreateInstance("Alt");

        await using var manager = Manager();
        await manager.StartAsync();

        (await StartAndCheck(manager, path, "Alt")).ShouldBe(StartRefusal.None);

        _tokens.Fuer.ShouldContain(Path.GetFullPath(path));
    }

    [Fact]
    public async Task A_failed_token_fetch_does_not_prevent_the_start()
    {
        // Monitoring that only runs with perfect permissions is silent exactly
        // when you need it.
        _tokens.Erfolg = false;

        var path = _directories.CreateInstance("Alt");

        await using var manager = Manager();
        await manager.StartAsync();

        (await StartAndCheck(manager, path, "Alt")).ShouldBe(StartRefusal.None);
    }

    // ─── The verdict ───

    /// <summary>
    /// The only case in which the manager stops instances on its own initiative.
    ///
    /// <para>A server outage explicitly does not belong here — the SDK separates verdict and
    /// failed attempt, and only the verdict arrives here. Otherwise a
    /// network problem would take all installations down at the same time.</para>
    /// </summary>
    [Fact]
    public async Task A_license_revocation_pauses_all_running_instances()
    {
        var path = _directories.CreateInstance("Alpha");

        await using var manager = Manager();
        await manager.StartAsync();
        await StartAndCheck(manager, path, "Alpha");

        manager.Instances.Instances.Single().Condition.ShouldBe(InstanceCondition.Running);

        // The instance ends upon the request.
        var process = _launcher.Last!;
        _channel.OnStopRequest = _ => { _channel.AnswerFor[Path.GetFullPath(path)] = null; process.Exit(0); };

        await manager.HandleLicenseRevokedAsync("Widerrufen durch den Hersteller");

        manager.LicenseHolds.ShouldBeFalse();
        manager.Instances.Instances.Single().Condition.ShouldBe(InstanceCondition.Stopped);
        _channel.StopRequests.ShouldBe(1, "Erst bitten, nicht schiessen.");
    }

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

        await using var manager = Manager();
        await manager.StartAsync();

        await manager.HandleLicenseRevokedAsync("Widerrufen");

        (await manager.StartInstanceAsync(path)).ShouldBe(StartRefusal.NoLicense);

        _launcher.Starts.ShouldBe(0);
        _tokens.Fuer.ShouldBeEmpty("Ohne Lizenz wird auch kein Token verbraucht.");
    }

    // ─── When the start fails ───

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

        await using var manager = Manager();
        await manager.StartAsync();

        // The process runs, the instance never reports in.
        (await manager.StartInstanceAsync(path)).ShouldBe(StartRefusal.DidNotStart);
    }
}