Files
ClaudeDo/src/ClaudeDo.Worker/Program.cs
mika kuns d094a21e09 feat(planning): prevent orphaned subtasks via guards + startup repair
Three coordinated guards close the orphan-creation paths:

- CreateChildAsync refuses when the parent is not in a planning phase.
- DiscardPlanningAsync now returns a structured DiscardPlanningOutcome
  and refuses when children are queued or running; callers can opt into
  auto-dequeuing queued kids via dequeueQueuedChildren=true. Terminal
  children (Done/Failed/Cancelled) are promoted to top-level instead of
  becoming orphans when the parent's PlanningPhase is reset.
- OrphanRecovery hosted service clears ParentTaskId on any rows whose
  parent is missing or no longer in a planning phase on worker startup,
  mirroring the StaleTaskRecovery pattern.

UI surfaces the block reason: a confirm dialog offers to dequeue queued
children and retry; a running-children block is shown as a hard error
asking the user to cancel first.

WorkerClient now negotiates the JsonStringEnumConverter so the
DiscardPlanningResult enum round-trips correctly over SignalR.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-18 16:02:15 +02:00

209 lines
8.9 KiB
C#

using ClaudeDo.Data;
using ClaudeDo.Data.Git;
using ClaudeDo.Data.Repositories;
using ClaudeDo.Worker.Agents;
using ClaudeDo.Worker.Config;
using ClaudeDo.Worker.External;
using ClaudeDo.Worker.Hub;
using ClaudeDo.Worker.Lifecycle;
using ClaudeDo.Worker.Planning;
using ClaudeDo.Worker.Queue;
using ClaudeDo.Worker.Runner;
using ClaudeDo.Worker.State;
using ClaudeDo.Worker.Prime;
using ClaudeDo.Worker.Worktrees;
using Microsoft.EntityFrameworkCore;
var cfg = WorkerConfig.Load();
var builder = WebApplication.CreateBuilder(args);
// When launched by the Windows SCM, speak the Service Control Protocol so SCM
// doesn't think we crashed (~30s timeout). No-op when running interactively.
builder.Host.UseWindowsService(o => o.ServiceName = "ClaudeDoWorker");
builder.Services.AddDbContextFactory<ClaudeDoDbContext>(opt =>
opt.UseSqlite($"Data Source={cfg.DbPath}"));
builder.Services.AddSingleton(cfg);
builder.Services.AddHostedService<StaleTaskRecovery>();
builder.Services.AddHostedService<OrphanRecovery>();
builder.Services.AddSignalR().AddJsonProtocol(options =>
{
options.PayloadSerializerOptions.Converters.Add(new System.Text.Json.Serialization.JsonStringEnumConverter());
});
// Runner stack.
builder.Services.AddSingleton<IClaudeProcess, ClaudeProcess>();
builder.Services.AddSingleton<HubBroadcaster>();
builder.Services.AddSingleton<GitService>();
builder.Services.AddSingleton<WorktreeManager>();
builder.Services.AddSingleton<ClaudeArgsBuilder>();
builder.Services.AddSingleton<TaskRunner>();
builder.Services.AddSingleton<WorktreeMaintenanceService>();
builder.Services.AddSingleton<TaskResetService>();
builder.Services.AddSingleton<TaskMergeService>();
builder.Services.AddSingleton<PlanningAggregator>();
builder.Services.AddSingleton<PlanningMergeOrchestrator>();
builder.Services.AddSingleton<PlanningChainCoordinator>();
// Queue dispatch primitives. QueueWaker holds the wake semaphore; the queue picker
// performs atomic Queued→Running claim. Both injected into the state service so it
// can wake the dispatcher without depending on QueueService directly.
builder.Services.AddSingleton<QueueWaker>();
builder.Services.AddSingleton<IQueueWaker>(sp => sp.GetRequiredService<QueueWaker>());
builder.Services.AddSingleton<IQueuePicker, QueuePicker>();
builder.Services.AddSingleton<Func<ITaskStateService>>(sp => () => sp.GetRequiredService<ITaskStateService>());
builder.Services.AddSingleton<ITaskStateService>(sp => new TaskStateService(
sp.GetRequiredService<IDbContextFactory<ClaudeDoDbContext>>(),
sp.GetRequiredService<HubBroadcaster>(),
sp.GetRequiredService<IQueueWaker>(),
sp.GetRequiredService<PlanningChainCoordinator>(),
sp.GetRequiredService<ILogger<TaskStateService>>()));
// Agent file management.
var agentsDir = Path.Combine(ClaudeDo.Data.Paths.AppDataRoot(), "agents");
Directory.CreateDirectory(agentsDir);
builder.Services.AddSingleton(new AgentFileService(agentsDir));
var defaultAgentsBundleDir = Path.Combine(AppContext.BaseDirectory, "DefaultAgents");
builder.Services.AddSingleton(sp => new DefaultAgentSeeder(
defaultAgentsBundleDir,
agentsDir,
sp.GetService<Microsoft.Extensions.Logging.ILogger<DefaultAgentSeeder>>()));
// Override slot owns RunNow / ContinueTask. Queue slot is the BackgroundService.
builder.Services.AddSingleton<OverrideSlotService>();
// Prime Claude
builder.Services.AddSingleton<IPrimeClock, PrimeClock>();
builder.Services.AddSingleton<PrimeScheduleSignal>();
builder.Services.AddSingleton<IPrimeScheduleSignal>(sp => sp.GetRequiredService<PrimeScheduleSignal>());
builder.Services.AddSingleton<IPrimeRunner, PrimeRunner>();
builder.Services.AddSingleton(PrimeSchedulerOptions.Default);
builder.Services.AddSingleton<IPrimeBroadcaster>(sp => sp.GetRequiredService<HubBroadcaster>());
builder.Services.AddHostedService<PrimeScheduler>();
// QueueService: singleton + hosted service (same instance).
builder.Services.AddSingleton<QueueService>();
builder.Services.AddHostedService(sp => sp.GetRequiredService<QueueService>());
// Planning session services.
var planningSessionsDir = Path.Combine(
Environment.GetFolderPath(Environment.SpecialFolder.UserProfile),
".todo-app", "planning-sessions");
builder.Services.AddSingleton(sp =>
new PlanningSessionManager(
sp.GetRequiredService<IDbContextFactory<ClaudeDoDbContext>>(),
sp.GetRequiredService<GitService>(),
cfg,
sp.GetRequiredService<ITaskStateService>(),
sp.GetRequiredService<PlanningChainCoordinator>(),
planningSessionsDir));
builder.Services.AddSingleton<IPlanningTerminalLauncher>(sp =>
new WindowsTerminalPlanningLauncher("wt.exe", cfg.ClaudeBin));
builder.Services.AddHttpContextAccessor();
builder.Services.AddScoped<PlanningMcpContextAccessor>();
builder.Services.AddScoped<ClaudeDoDbContext>(sp =>
sp.GetRequiredService<IDbContextFactory<ClaudeDoDbContext>>().CreateDbContext());
builder.Services.AddScoped<TaskRepository>();
builder.Services.AddScoped<ListRepository>();
builder.Services.AddScoped<PlanningMcpService>();
builder.Services.AddMcpServer()
.WithHttpTransport()
.WithTools<PlanningMcpService>();
// Loopback-only bind. Firewall is irrelevant for 127.0.0.1.
builder.WebHost.UseUrls($"http://127.0.0.1:{cfg.SignalRPort}");
var app = builder.Build();
using (var scope = app.Services.CreateScope())
{
ClaudeDoDbContext.MigrateAndConfigure(
scope.ServiceProvider.GetRequiredService<ClaudeDoDbContext>());
}
try
{
var seeder = app.Services.GetRequiredService<DefaultAgentSeeder>();
var seedResult = await seeder.SeedMissingAsync();
app.Logger.LogInformation(
"Default agents seeded: {Copied} copied, {Skipped} already present",
seedResult.Copied, seedResult.Skipped);
}
catch (Exception ex)
{
app.Logger.LogWarning(ex, "Default agent seeding failed");
}
app.UseMiddleware<PlanningTokenAuthMiddleware>();
app.MapHub<WorkerHub>("/hub");
app.MapMcp("/mcp");
// Claude CLI preflight: fail fast if the configured binary is unreachable or non-zero.
// Skippable via CLAUDEDO_SKIP_CLI_PREFLIGHT=1 for environments without the CLI (e.g. tests).
if (Environment.GetEnvironmentVariable("CLAUDEDO_SKIP_CLI_PREFLIGHT") != "1")
{
var preflight = await ClaudeCliPreflight.CheckAsync(cfg.ClaudeBin);
if (!preflight.Ok)
{
app.Logger.LogCritical(
"Claude CLI preflight failed (bin: '{Bin}', exit: {Exit}): {Error}. " +
"Fix `claude_bin` in worker.config.json or set CLAUDEDO_SKIP_CLI_PREFLIGHT=1 to bypass.",
cfg.ClaudeBin, preflight.ExitCode, preflight.Error);
Environment.Exit(1);
}
app.Logger.LogInformation("Claude CLI preflight OK: {Version}", preflight.Version);
}
app.Logger.LogInformation("ClaudeDo.Worker listening on http://127.0.0.1:{Port} (db: {Db})",
cfg.SignalRPort, cfg.DbPath);
// Build the external MCP endpoint as a separate WebApplication on its own port.
// Rationale: ModelContextProtocol.AspNetCore registers one server per DI container,
// so we need a second app to expose a different tool set under different auth.
// Shared singletons (QueueService, HubBroadcaster, WorkerConfig, db factory) are
// injected by instance so both apps operate on the same runtime state.
WebApplication? externalApp = null;
if (cfg.ExternalMcpPort > 0)
{
var externalBuilder = WebApplication.CreateBuilder();
externalBuilder.Services.AddSingleton(cfg);
externalBuilder.Services.AddSingleton(app.Services.GetRequiredService<HubBroadcaster>());
externalBuilder.Services.AddSingleton(app.Services.GetRequiredService<QueueService>());
externalBuilder.Services.AddSingleton(app.Services.GetRequiredService<OverrideSlotService>());
externalBuilder.Services.AddSingleton(app.Services.GetRequiredService<IDbContextFactory<ClaudeDoDbContext>>());
externalBuilder.Services.AddSingleton(app.Services.GetRequiredService<ITaskStateService>());
externalBuilder.Services.AddSingleton(app.Services.GetRequiredService<IQueueWaker>());
externalBuilder.Services.AddScoped<ClaudeDoDbContext>(sp =>
sp.GetRequiredService<IDbContextFactory<ClaudeDoDbContext>>().CreateDbContext());
externalBuilder.Services.AddScoped<TaskRepository>();
externalBuilder.Services.AddScoped<ListRepository>();
externalBuilder.Services.AddScoped<TagRepository>();
externalBuilder.Services.AddScoped<ExternalMcpService>();
externalBuilder.Services.AddMcpServer()
.WithHttpTransport()
.WithTools<ExternalMcpService>();
externalBuilder.WebHost.UseUrls($"http://127.0.0.1:{cfg.ExternalMcpPort}");
externalApp = externalBuilder.Build();
externalApp.UseMiddleware<ExternalMcpAuthMiddleware>();
externalApp.MapMcp("/mcp");
externalApp.Logger.LogInformation(
"ClaudeDo.Worker external MCP listening on http://127.0.0.1:{Port} (auth: {Auth})",
cfg.ExternalMcpPort,
string.IsNullOrEmpty(cfg.ExternalMcpApiKey) ? "loopback-only" : "X-ClaudeDo-Key");
}
if (externalApp is null)
{
await app.RunAsync();
}
else
{
await Task.WhenAll(app.RunAsync(), externalApp.RunAsync());
}