Rund 2.600 Kommentarblöcke in 310 C#-Dateien übersetzt, Begründungen vollständig erhalten: blockweise gleiche Anzahl wie vorher, kein Block unter 60 % seiner Länge, der Code außerhalb der Kommentare per Tokenizer byte-identisch geprüft. Zitate deutscher Oberflächentexte und Testdaten bleiben deutsch. WebUI: Kommentare in JS/CSS/HTML übersetzt, JSDoc-@param an die in 4b umbenannten Parameter angeglichen. Dabei behoben: - 4b hatte "tag" (HTML-Tag) als deutsches "Tag" gelesen und zu "dayValue" gemacht (makeElement, svg, Tagesübersicht) - zurück auf tag/day. - Aufgaben "InProgress" bekamen die Plakettenart "laufend", für die es nie eine CSS-Klasse gab - jetzt "success" wie laufende Instanzen. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
444 lines
16 KiB
C#
444 lines
16 KiB
C#
using System.Text.Json;
|
|
using ClawdDotNet.Core.Config;
|
|
using ClawdDotNet.Core.Instances;
|
|
using Microsoft.Extensions.Logging.Abstractions;
|
|
using Shouldly;
|
|
|
|
namespace ClawdDotNet.Manager.Tests;
|
|
|
|
/// <summary>
|
|
/// "Autostart by manager" — the switch that keeps an instance running (#528).
|
|
///
|
|
/// <para><b>Two things in one switch, because it is the same promise:</b> the manager
|
|
/// starts the instance when coming up, and it starts it again after a crash. Without
|
|
/// the switch it touches nothing — a manager that boots everything it finds after a restart of the machine
|
|
/// also starts what someone stopped on purpose.</para>
|
|
///
|
|
/// <para><b>And the circuit breaker is the actual content.</b> Booting again
|
|
/// is the easy half; the hard one is recognizing when to stop.</para>
|
|
/// </summary>
|
|
public sealed class AutostartTests : IAsyncDisposable
|
|
{
|
|
private readonly string _root = Path.Combine(
|
|
Path.GetTempPath(), "clawd-tests", "auto-" + Guid.NewGuid().ToString("N")[..8]);
|
|
|
|
private readonly FakeLauncher _launcher = new();
|
|
private readonly FakeChannel _kanal = new();
|
|
private readonly FakeTokenSource _tokens = new();
|
|
private readonly InstanceDirectoryManager _verzeichnisse;
|
|
|
|
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 AutostartTests()
|
|
{
|
|
Directory.CreateDirectory(_root);
|
|
_verzeichnisse = new InstanceDirectoryManager(Path.Combine(_root, "Instances"));
|
|
|
|
// The socket only answers if a process is really running - otherwise an
|
|
// answer stored in advance would mean "already running", and the manager would take over
|
|
// instead of starting.
|
|
_kanal.Launcher = _launcher;
|
|
}
|
|
|
|
public async ValueTask DisposeAsync()
|
|
{
|
|
await Task.CompletedTask;
|
|
try { Directory.Delete(_root, recursive: true); }
|
|
catch { /* A leftover test folder does not turn a run red. */ }
|
|
}
|
|
|
|
private ManagerCore Manager(RestartPolicy? policy = null)
|
|
{
|
|
var register = new InstanceRegistry(
|
|
_verzeichnisse, _launcher, _kanal, NullLoggerFactory.Instance, Options);
|
|
|
|
return new ManagerCore(
|
|
new ManagerSettingsStore(_root), register, _tokens, NullLoggerFactory.Instance,
|
|
restartPolicy: policy);
|
|
}
|
|
|
|
/// <summary>Creates an instance and sets the switch in its instance file.</summary>
|
|
private string Instanz(string name, bool autostart)
|
|
{
|
|
var pfad = _verzeichnisse.CreateInstance(name);
|
|
var datei = Path.Combine(pfad, "InstanceSettings.json");
|
|
|
|
var config = JsonSerializer.Deserialize<InstanceConfig>(File.ReadAllText(datei))!;
|
|
config.AutostartByManager = autostart;
|
|
File.WriteAllText(datei, JsonSerializer.Serialize(config));
|
|
|
|
_kanal.AnswerFor[Path.GetFullPath(pfad)] = new InstanceIdentity("i-" + name, name, "0.1.2");
|
|
|
|
return pfad;
|
|
}
|
|
|
|
private static SupervisionEvent Absturz(string pfad, string name) => new(
|
|
InstanceEventKind.Crashed, Path.GetFullPath(pfad), name, 4711, 137,
|
|
"unerwartet beendet", DateTimeOffset.UtcNow);
|
|
|
|
// ─── When coming up ───
|
|
|
|
[Fact]
|
|
public async Task Eine_Instanz_mit_Autostart_wird_beim_Hochfahren_gestartet()
|
|
{
|
|
Instanz("Alpha", autostart: true);
|
|
|
|
await using var manager = Manager();
|
|
await manager.StartAsync();
|
|
|
|
_launcher.Starts.ShouldBe(1);
|
|
manager.Instances.Instances.Single().Condition.ShouldBe(InstanceCondition.Running);
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Ohne_Autostart_bleibt_sie_aus()
|
|
{
|
|
Instanz("Alpha", autostart: false);
|
|
|
|
await using var manager = Manager();
|
|
await manager.StartAsync();
|
|
|
|
_launcher.Starts.ShouldBe(0);
|
|
manager.Instances.Instances.Single().Condition.ShouldBe(InstanceCondition.Stopped);
|
|
}
|
|
|
|
/// <summary>
|
|
/// What is already running is not started again: a second process on the same
|
|
/// <c>state.db</c> would be the opposite of helpful.
|
|
/// </summary>
|
|
[Fact]
|
|
public async Task Was_schon_laeuft_wird_nicht_noch_einmal_gestartet()
|
|
{
|
|
var pfad = Instanz("Alpha", autostart: true);
|
|
|
|
await using (var erster = Manager())
|
|
await erster.StartAsync();
|
|
|
|
_launcher.Starts.ShouldBe(1);
|
|
|
|
await using var zweiter = Manager();
|
|
await zweiter.StartAsync();
|
|
|
|
_launcher.Starts.ShouldBe(1, "Uebernommen, nicht neu gestartet.");
|
|
zweiter.Instances.Instances.Single().ProcessId.ShouldBe(_launcher.Last!.Id);
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Der_Schalter_steht_in_der_Liste()
|
|
{
|
|
Instanz("Mit", autostart: true);
|
|
Instanz("Ohne", autostart: false);
|
|
|
|
await using var manager = Manager();
|
|
await manager.StartAsync();
|
|
|
|
manager.Instances.Instances.Single(i => i.Name == "Mit").AutostartByManager.ShouldBeTrue();
|
|
manager.Instances.Instances.Single(i => i.Name == "Ohne").AutostartByManager.ShouldBeFalse();
|
|
}
|
|
|
|
// ─── After a crash ───
|
|
|
|
[Fact]
|
|
public async Task Nach_einem_Absturz_wird_wieder_gestartet()
|
|
{
|
|
var pfad = Instanz("Alpha", autostart: true);
|
|
|
|
await using var manager = Manager();
|
|
await manager.StartAsync();
|
|
|
|
var vorher = _launcher.Starts;
|
|
|
|
// The process is gone; supervision has reported it.
|
|
_launcher.Last!.Exit(137);
|
|
await manager.RestartAfterCrashAsync(Absturz(pfad, "Alpha"));
|
|
|
|
_launcher.Starts.ShouldBe(vorher + 1);
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Ohne_Autostart_wird_nach_einem_Absturz_nichts_getan()
|
|
{
|
|
var pfad = Instanz("Alpha", autostart: false);
|
|
|
|
await using var manager = Manager();
|
|
await manager.StartAsync();
|
|
await manager.StartInstanceAsync(pfad);
|
|
|
|
_launcher.Last!.Exit(137);
|
|
await manager.RestartAfterCrashAsync(Absturz(pfad, "Alpha"));
|
|
|
|
_launcher.Starts.ShouldBe(1, "Nur der Start von Hand, kein Neustart.");
|
|
}
|
|
|
|
/// <summary>
|
|
/// A silent instance is ended and restarted — but only after the
|
|
/// old process is provably gone.
|
|
///
|
|
/// <para><b>The case is really produced here and not claimed.</b> The
|
|
/// process lives on, only the socket is silent — exactly the situation from #529. A test
|
|
/// that instead builds the event by hand and calls the restart method checks
|
|
/// the half that works anyway, and leaves out the dangerous one: that before
|
|
/// the start nobody sits on the database any more.</para>
|
|
/// </summary>
|
|
[Fact]
|
|
public async Task Eine_verstummte_Instanz_wird_erst_beendet_und_dann_neu_gestartet()
|
|
{
|
|
var pfad = Instanz("Alpha", autostart: true);
|
|
|
|
await using var manager = Manager();
|
|
await manager.StartAsync();
|
|
|
|
var alt = _launcher.Last.ShouldNotBeNull();
|
|
_launcher.Starts.ShouldBe(1);
|
|
|
|
// From now on it hangs: the process lives, the socket no longer answers.
|
|
_kanal.Verstummen(pfad);
|
|
|
|
var aufsicht = manager.Instances.SupervisorFor(pfad).ShouldNotBeNull();
|
|
|
|
// Three passes until the alarm - nothing is done before that.
|
|
for (var i = 0; i < 3; i++) await aufsicht.CheckAsync();
|
|
|
|
aufsicht.Condition.ShouldBe(InstanceCondition.Silent);
|
|
alt.WasKilled.ShouldBeFalse("Vor Ablauf der Frist wird nicht abgeschossen.");
|
|
|
|
// The deadline expires (SilentKillGrace: 300ms).
|
|
await Task.Delay(400);
|
|
await aufsicht.CheckAsync();
|
|
|
|
alt.WasKilled.ShouldBeTrue("Nach der Frist wird der haengende Prozess beendet.");
|
|
|
|
// The crash path restarts - and the new process is a different one.
|
|
await Warte(() => _launcher.Starts >= 2);
|
|
|
|
_launcher.Last.ShouldNotBeNull().Id.ShouldNotBe(alt.Id);
|
|
alt.HasExited.ShouldBeTrue("Der alte Prozess darf beim Start nicht mehr leben.");
|
|
}
|
|
|
|
/// <summary>
|
|
/// An instance that is cleaning up is not killed — even if it takes a while.
|
|
///
|
|
/// <para>It does answer; it is doing exactly what it was asked to do. The difference
|
|
/// between "hangs" and "cleans up" is the whole reason why the pulse says more than
|
|
/// "someone answered".</para>
|
|
/// </summary>
|
|
[Fact]
|
|
public async Task Wer_aufraeumt_gilt_nicht_als_verstummt()
|
|
{
|
|
var pfad = Instanz("Alpha", autostart: true);
|
|
|
|
await using var manager = Manager();
|
|
await manager.StartAsync();
|
|
|
|
var prozess = _launcher.Last.ShouldNotBeNull();
|
|
|
|
_kanal.AnswerFor[Path.GetFullPath(pfad)] = new InstanceIdentity(
|
|
"i-Alpha", "Alpha", "0.1.2",
|
|
ShutdownPhase.Draining, OpenWork: 2, OpenSummary: "Agentenläufe: 2");
|
|
|
|
var aufsicht = manager.Instances.SupervisorFor(pfad).ShouldNotBeNull();
|
|
|
|
for (var i = 0; i < 5; i++) await aufsicht.CheckAsync();
|
|
|
|
aufsicht.Condition.ShouldBe(InstanceCondition.Running);
|
|
prozess.WasKilled.ShouldBeFalse();
|
|
_launcher.Starts.ShouldBe(1);
|
|
}
|
|
|
|
private static async Task Warte(Func<bool> bedingung)
|
|
{
|
|
for (var i = 0; i < 100 && !bedingung(); i++)
|
|
await Task.Delay(20);
|
|
}
|
|
|
|
/// <summary>
|
|
/// After the limit the manager pauses.
|
|
///
|
|
/// <para>What fails because of a broken configuration fails the same way on the tenth attempt
|
|
/// — then instead of a dead instance you have a dead instance, a full
|
|
/// log and a bug tracker full of repetitions.</para>
|
|
/// </summary>
|
|
[Fact]
|
|
public async Task Nach_drei_Neustarts_im_Zeitfenster_setzt_der_Manager_aus()
|
|
{
|
|
var pfad = Instanz("Wackelig", autostart: true);
|
|
|
|
await using var manager = Manager(new RestartPolicy(MaxRestarts: 3, Window: TimeSpan.FromMinutes(10)));
|
|
await manager.StartAsync();
|
|
|
|
var nachStart = _launcher.Starts;
|
|
|
|
for (var i = 0; i < 6; i++)
|
|
{
|
|
_launcher.Last!.Exit(137);
|
|
await manager.RestartAfterCrashAsync(Absturz(pfad, "Wackelig"));
|
|
}
|
|
|
|
(_launcher.Starts - nachStart).ShouldBe(3, "Drei Versuche, dann Schluss.");
|
|
}
|
|
|
|
/// <summary>
|
|
/// Whoever starts by hand has either changed something or knows what they are doing. Charging them with
|
|
/// the earlier attempts would mean punishing them for the crash.
|
|
/// </summary>
|
|
[Fact]
|
|
public async Task Ein_Start_von_Hand_raeumt_den_Zaehler_ab()
|
|
{
|
|
var pfad = Instanz("Wackelig", autostart: true);
|
|
|
|
await using var manager = Manager(new RestartPolicy(3, TimeSpan.FromMinutes(10)));
|
|
await manager.StartAsync();
|
|
|
|
for (var i = 0; i < 5; i++)
|
|
{
|
|
_launcher.Last!.Exit(137);
|
|
await manager.RestartAfterCrashAsync(Absturz(pfad, "Wackelig"));
|
|
}
|
|
|
|
var erschoepft = _launcher.Starts;
|
|
|
|
// A new manager starts at zero - just like a start by hand.
|
|
await using var frisch = Manager(new RestartPolicy(3, TimeSpan.FromMinutes(10)));
|
|
await frisch.StartAsync();
|
|
|
|
_launcher.Last!.Exit(137);
|
|
await frisch.RestartAfterCrashAsync(Absturz(pfad, "Wackelig"));
|
|
|
|
_launcher.Starts.ShouldBeGreaterThan(erschoepft);
|
|
}
|
|
}
|
|
|
|
/// <summary>The circuit breaker on its own.</summary>
|
|
public sealed class RestartLedgerTests
|
|
{
|
|
private sealed class Uhr(DateTimeOffset jetzt) : TimeProvider
|
|
{
|
|
public DateTimeOffset Jetzt { get; set; } = jetzt;
|
|
public override DateTimeOffset GetUtcNow() => Jetzt;
|
|
}
|
|
|
|
[Fact]
|
|
public void Bis_zur_Grenze_ja_danach_nein()
|
|
{
|
|
var buch = new RestartLedger(new RestartPolicy(3, TimeSpan.FromMinutes(10)));
|
|
|
|
buch.TryRecord("/a", out _).ShouldBeTrue();
|
|
buch.TryRecord("/a", out _).ShouldBeTrue();
|
|
buch.TryRecord("/a", out var dritter).ShouldBeTrue();
|
|
dritter.ShouldBe(3);
|
|
|
|
buch.TryRecord("/a", out _).ShouldBeFalse();
|
|
}
|
|
|
|
[Fact]
|
|
public void Jede_Instanz_hat_ihr_eigenes_Konto()
|
|
{
|
|
var buch = new RestartLedger(new RestartPolicy(1, TimeSpan.FromMinutes(10)));
|
|
|
|
buch.TryRecord("/a", out _).ShouldBeTrue();
|
|
buch.TryRecord("/a", out _).ShouldBeFalse();
|
|
|
|
buch.TryRecord("/b", out _).ShouldBeTrue("Die eine Instanz sperrt die andere nicht.");
|
|
}
|
|
|
|
/// <summary>
|
|
/// The time window moves: an instance that crashes once a month and comes back
|
|
/// up never runs into the limit — one that falls every minute does immediately.
|
|
/// </summary>
|
|
[Fact]
|
|
public void Alte_Versuche_fallen_aus_dem_Fenster()
|
|
{
|
|
var uhr = new Uhr(DateTimeOffset.UnixEpoch);
|
|
var buch = new RestartLedger(new RestartPolicy(2, TimeSpan.FromMinutes(10)), uhr);
|
|
|
|
buch.TryRecord("/a", out _).ShouldBeTrue();
|
|
buch.TryRecord("/a", out _).ShouldBeTrue();
|
|
buch.TryRecord("/a", out _).ShouldBeFalse();
|
|
|
|
uhr.Jetzt = uhr.Jetzt.AddMinutes(11);
|
|
|
|
buch.TryRecord("/a", out var wieder).ShouldBeTrue();
|
|
wieder.ShouldBe(1, "Die alten Versuche zaehlen nicht mehr mit.");
|
|
}
|
|
|
|
[Fact]
|
|
public void Vergessen_setzt_zurueck()
|
|
{
|
|
var buch = new RestartLedger(new RestartPolicy(1, TimeSpan.FromMinutes(10)));
|
|
|
|
buch.TryRecord("/a", out _).ShouldBeTrue();
|
|
buch.TryRecord("/a", out _).ShouldBeFalse();
|
|
|
|
buch.Forget("/a");
|
|
|
|
buch.TryRecord("/a", out _).ShouldBeTrue();
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// A start that meets an outdated note.
|
|
///
|
|
/// <para><b>The defect that only the autostart tests revealed.</b> Between two
|
|
/// supervision passes there are seconds, and in them an instance can break away. Whoever
|
|
/// asked to "start" in this gap got "already running" — a success with nothing
|
|
/// behind it. The next pass then reported a crash nobody expected any more.</para>
|
|
/// </summary>
|
|
public sealed class StaleStateTests : IAsyncDisposable
|
|
{
|
|
private readonly string _root = Path.Combine(
|
|
Path.GetTempPath(), "clawd-tests", "stale-" + Guid.NewGuid().ToString("N")[..8]);
|
|
|
|
private readonly FakeLauncher _launcher = new();
|
|
private readonly FakeChannel _kanal = new();
|
|
|
|
public StaleStateTests()
|
|
{
|
|
Directory.CreateDirectory(_root);
|
|
_kanal.Launcher = _launcher;
|
|
_kanal.Answer = new InstanceIdentity("i-1", "Alpha", "0.1.2");
|
|
}
|
|
|
|
public async ValueTask DisposeAsync()
|
|
{
|
|
await Task.CompletedTask;
|
|
try { Directory.Delete(_root, recursive: true); }
|
|
catch { /* A leftover test folder does not turn a run red. */ }
|
|
}
|
|
|
|
[Fact]
|
|
public async Task Ein_Start_prueft_nach_statt_dem_eigenen_Zustand_zu_glauben()
|
|
{
|
|
var dir = Path.Combine(_root, "Instance-Alpha");
|
|
Directory.CreateDirectory(dir);
|
|
|
|
await using var aufsicht = new InstanceSupervisor(
|
|
dir, _launcher, _kanal, NullLogger.Instance,
|
|
new SupervisionOptions(
|
|
TimeSpan.FromSeconds(30), // deliberately sluggish: no pass in between
|
|
TimeSpan.FromMilliseconds(600),
|
|
3,
|
|
TimeSpan.FromMilliseconds(300),
|
|
TimeSpan.FromMilliseconds(300),
|
|
TimeSpan.FromSeconds(2)));
|
|
|
|
(await aufsicht.StartAsync()).ShouldBeTrue();
|
|
aufsicht.Condition.ShouldBe(InstanceCondition.Running);
|
|
|
|
// The instance breaks away. Supervision has not noticed yet - the next
|
|
// pass would only come in thirty seconds.
|
|
_launcher.Last!.Exit(137);
|
|
aufsicht.Condition.ShouldBe(InstanceCondition.Running, "Noch nichts bemerkt.");
|
|
|
|
(await aufsicht.StartAsync()).ShouldBeTrue();
|
|
|
|
_launcher.Starts.ShouldBe(2, "Es wurde wirklich gestartet, nicht nur zugestimmt.");
|
|
aufsicht.Condition.ShouldBe(InstanceCondition.Running);
|
|
}
|
|
}
|