fix(worker): claim Running before creating run resources

TaskRunner.RunAsync created the worktree (PrepareRunDirectoryAsync)
before claiming Running via StartRunningAsync. RunNow dispatches by
task id with no atomic claim of their own, so a Queued task racing
the queue picker's atomic SQL claim could hit WorktreeManager's
branch-collision self-heal, which force-removes and recreates the
winner's live worktree mid-run.

Move the claim before any resource creation and bail out immediately
when it's rejected. OverrideSlotService.RunNow also fast-rejects a
task already Running in the DB (defense in depth). RunCancellationRegistry
now refuses (and logs) a double registration instead of silently
overwriting the first CTS, so a losing dispatch's cleanup can no
longer unregister the winner's cancellation token.
This commit is contained in:
mika kuns
2026-08-06 13:33:23 +02:00
parent 0d1e3b9a6f
commit 774f9d3d13
10 changed files with 298 additions and 41 deletions
@@ -1,4 +1,5 @@
using ClaudeDo.Worker.Queue;
using Microsoft.Extensions.Logging.Abstractions;
namespace ClaudeDo.Worker.Tests.Queue;
@@ -7,9 +8,9 @@ public sealed class RunCancellationRegistryTests
[Fact]
public void TryCancel_RegisteredTask_CancelsAndReturnsTrue()
{
var sut = new RunCancellationRegistry();
var sut = new RunCancellationRegistry(NullLogger<RunCancellationRegistry>.Instance);
using var cts = new CancellationTokenSource();
sut.Register("t1", cts);
Assert.True(sut.Register("t1", cts));
Assert.True(sut.TryCancel("t1"));
Assert.True(cts.IsCancellationRequested);
@@ -18,30 +19,50 @@ public sealed class RunCancellationRegistryTests
[Fact]
public void TryCancel_UnknownTask_ReturnsFalse()
{
var sut = new RunCancellationRegistry();
var sut = new RunCancellationRegistry(NullLogger<RunCancellationRegistry>.Instance);
Assert.False(sut.TryCancel("nope"));
}
[Fact]
public void Register_SecondCallForSameTask_RefusesAndKeepsFirstRegistration()
{
// A double-dispatch (e.g. RunNow racing the queue picker for the same task) must not
// let the second registration silently clobber the first — that would leave CancelAsync
// pointing at the wrong (or already-finished) CTS.
var sut = new RunCancellationRegistry(NullLogger<RunCancellationRegistry>.Instance);
using var first = new CancellationTokenSource();
using var second = new CancellationTokenSource();
Assert.True(sut.Register("t1", first));
Assert.False(sut.Register("t1", second));
Assert.True(sut.TryCancel("t1"));
Assert.True(first.IsCancellationRequested);
Assert.False(second.IsCancellationRequested);
}
[Fact]
public void Unregister_RemovesOnlyTheGivenRegistration()
{
var sut = new RunCancellationRegistry();
using var stale = new CancellationTokenSource();
using var current = new CancellationTokenSource();
var sut = new RunCancellationRegistry(NullLogger<RunCancellationRegistry>.Instance);
using var first = new CancellationTokenSource();
sut.Register("t1", stale);
sut.Register("t1", current); // re-run replaced the registration
sut.Unregister("t1", stale); // late cleanup of the old slot must not evict the new one
sut.Register("t1", first);
sut.Unregister("t1", first);
// Slot is free again once the run that held it cleans up properly, so a genuine
// sequential re-run can register a fresh CTS under the same task id.
using var second = new CancellationTokenSource();
Assert.True(sut.Register("t1", second));
Assert.True(sut.TryCancel("t1"));
Assert.True(current.IsCancellationRequested);
Assert.False(stale.IsCancellationRequested);
Assert.True(second.IsCancellationRequested);
}
[Fact]
public void TryCancel_DisposedCts_ReturnsFalse()
{
var sut = new RunCancellationRegistry();
var sut = new RunCancellationRegistry(NullLogger<RunCancellationRegistry>.Instance);
var cts = new CancellationTokenSource();
sut.Register("t1", cts);
cts.Dispose();