fix(backend): recreate the worktree when resuming an unarchived task (#2508)

Co-authored-by: Kandev Agent <agent@kandev.dev> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-11 00:19:40 +02:00
parent 6a093d8eae
commit a69a026741
4 changed files with 290 additions and 19 deletions
@@ -172,16 +172,41 @@ func isAgentAlreadyRunningError(err error) bool {
 	return err != nil && errors.Is(err, lifecycle.ErrAgentAlreadyRunning)
 }

-func validateSessionWorktrees(session *models.TaskSession) error {
+// missingSessionWorktreePaths returns the session's recorded worktree
+// directories that are no longer on disk.
+func missingSessionWorktreePaths(session *models.TaskSession) []string {
+	var missing []string
 	for _, wt := range session.Worktrees {
-		if wt.WorktreePath == "" {
+		if wt == nil || wt.WorktreePath == "" {
 			continue
 		}
 		if _, err := os.Stat(wt.WorktreePath); err != nil {
-			return fmt.Errorf("worktree path not found: %w", err)
+			missing = append(missing, wt.WorktreePath)
 		}
 	}
-	return nil
+	return missing
+}
+
+// noteMissingWorktreesBeforeResume records that a resume is about to go
+// through the worktree manager's recreate path.
+//
+// A missing directory is NOT a resume blocker. Archive cleanup deletes the
+// worktree directory while keeping the environment repository row and the git
+// branch (DestroyWorktree passes removeBranch=false), so *every* resume after
+// an unarchive observes exactly this shape. worktree.Manager.Create reuses the
+// stored record, restores the branch — locally, or by fetching it back from
+// origin — and rebuilds the directory at the same path, which is what the
+// recreate path was written for. Rejecting the resume here instead made
+// unarchive a dead end: the session could never be resumed and the workspace
+// was never restored, even though the work itself was still intact.
+func (s *Service) noteMissingWorktreesBeforeResume(sessionID string, session *models.TaskSession) {
+	missing := missingSessionWorktreePaths(session)
+	if len(missing) == 0 {
+		return
+	}
+	s.logger.Info("worktree directory missing on resume; it will be recreated from the stored branch",
+		zap.String("session_id", sessionID),
+		zap.Strings("paths", missing))
 }

 // EnqueueTask manually adds a task to the queue
@@ -1584,9 +1609,7 @@ func (s *Service) ResumeTaskSession(ctx context.Context, taskID, sessionID strin
 		// the record may already have been cleaned up before the user clicked Resume — allow it.
 		return nil, fmt.Errorf("session is not resumable: no executor record")
 	}
-	if err := validateSessionWorktrees(session); err != nil {
-		return nil, err
-	}
+	s.noteMissingWorktreesBeforeResume(sessionID, session)

 	// Completed sessions cannot be restarted — they require a new session.
 	// Failed and cancelled sessions keep the resume token so the relaunched
@@ -1970,9 +1993,7 @@ func (s *Service) attemptColdResume(
 		return false, fmt.Errorf("session is not resumable: no executor record (state: %s)", session.State)
 	}

-	if err := validateSessionWorktrees(session); err != nil {
-		return false, err
-	}
+	s.noteMissingWorktreesBeforeResume(sessionID, session)

 	// Use context.WithoutCancel to prevent WebSocket request timeout from canceling the resume.
 	// The lifecycle layer publishes events.AgentBootReady (handled by handleAgentBootReady)
@@ -2520,14 +2541,11 @@ func (s *Service) validateResumeEligibility(session *models.TaskSession, resp dt
 		return resp
 	}

-	// Check if worktree exists (if one was used)
-	if len(session.Worktrees) > 0 && session.Worktrees[0].WorktreePath != "" {
-		if _, err := os.Stat(session.Worktrees[0].WorktreePath); err != nil {
-			resp.Error = "worktree not found"
-			resp.IsResumable = false
-			return resp
-		}
-	}
+	// A missing worktree directory does not make the session unresumable.
+	// Archive cleanup deletes the directory and keeps the branch, so reporting
+	// IsResumable=false here would hide the resume affordance for every
+	// unarchived task — see noteMissingWorktreesBeforeResume. The resume
+	// itself recreates the directory through the worktree manager.

 	// Don't auto-resume sessions in error-recovery state.
 	if isErrorRecoveryState(session) {
@@ -0,0 +1,171 @@
+package orchestrator
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/kandev/kandev/internal/orchestrator/executor"
+	"github.com/kandev/kandev/internal/task/models"
+	v1 "github.com/kandev/kandev/pkg/api/v1"
+)
+
+// TestResumeTaskSession_RecreatesMissingWorktreeAfterUnarchive pins the
+// unarchive → resume path. Archiving a task deletes the worktree *directory*
+// but keeps the environment repository row and the git branch, so every
+// resume after an unarchive sees a recorded worktree path that no longer
+// exists on disk. Rejecting the resume there made unarchive a dead end — the
+// session could never be resumed even though the branch was still intact.
+// The resume must reach the launch instead, where the worktree manager
+// recreates the directory from the stored record.
+func TestResumeTaskSession_RecreatesMissingWorktreeAfterUnarchive(t *testing.T) {
+	ctx := context.Background()
+	repo := setupTestRepo(t)
+	// CANCELLED with the archive reason and no executors_running row: exactly
+	// what archive cleanup leaves behind for an unarchived task.
+	seedTaskAndSession(t, repo, "task1", "session1", models.TaskSessionStateCancelled)
+
+	session, err := repo.GetTaskSession(ctx, "session1")
+	if err != nil {
+		t.Fatalf("failed to load session: %v", err)
+	}
+	session.AgentProfileID = "profile1"
+	session.ErrorMessage = models.SessionArchiveCancelReason
+	session.TaskEnvironmentID = "env1"
+	if err := repo.UpdateTaskSession(ctx, session); err != nil {
+		t.Fatalf("failed to update session: %v", err)
+	}
+	if err := repo.CreateTaskEnvironment(ctx, &models.TaskEnvironment{
+		ID: "env1", TaskID: "task1", ExecutorType: "worktree",
+		WorkspacePath: t.TempDir(), Status: models.TaskEnvironmentStatusReady,
+	}); err != nil {
+		t.Fatalf("CreateTaskEnvironment: %v", err)
+	}
+	// The directory below is never created — the archive removed it.
+	missingPath := filepath.Join(t.TempDir(), "archived-task", "kandev")
+	if err := repo.CreateTaskEnvironmentRepo(ctx, &models.TaskEnvironmentRepo{
+		ID:                "ter1",
+		TaskEnvironmentID: "env1",
+		WorktreeID:        "wid1",
+		RepositoryID:      "repo1",
+		WorktreePath:      missingPath,
+		WorktreeBranch:    "feature/archived-work",
+	}); err != nil {
+		t.Fatalf("CreateTaskEnvironmentRepo: %v", err)
+	}
+	if _, statErr := os.Stat(missingPath); statErr == nil {
+		t.Fatalf("test setup is wrong: %s must not exist on disk", missingPath)
+	}
+
+	var launched atomic.Int32
+	agentMgr := &mockAgentManager{
+		repoForExecutionLookup: repo,
+		isAgentReadyFn:         func(_ context.Context, _ string) bool { return true },
+		launchAgentFunc: func(_ context.Context, req *executor.LaunchAgentRequest) (*executor.LaunchAgentResponse, error) {
+			launched.Add(1)
+			sess, getErr := repo.GetTaskSession(context.Background(), req.SessionID)
+			if getErr == nil && sess != nil {
+				sess.State = models.TaskSessionStateWaitingForInput
+				sess.UpdatedAt = time.Now().UTC()
+				_ = repo.UpdateTaskSession(context.Background(), sess)
+			}
+			return &executor.LaunchAgentResponse{AgentExecutionID: "exec-unarchived"}, nil
+		},
+	}
+
+	taskRepo := newMockTaskRepo()
+	taskRepo.tasks["task1"] = &v1.Task{ID: "task1", Title: "Test Task", State: v1.TaskStateInProgress}
+	svc := createTestServiceWithAgent(repo, newMockStepGetter(), taskRepo, agentMgr)
+	svc.executor = executor.NewExecutor(agentMgr, repo, testLogger(), executor.ExecutorConfig{})
+
+	exec, err := svc.ResumeTaskSession(ctx, "task1", "session1")
+	if err != nil {
+		t.Fatalf("resume of an unarchived session must not be blocked by the missing worktree directory: %v", err)
+	}
+	if exec == nil {
+		t.Fatal("ResumeTaskSession returned nil execution")
+	}
+	if launched.Load() == 0 {
+		t.Fatal("expected the resume to reach LaunchAgent, where the worktree is recreated")
+	}
+}
+
+// TestGetTaskSessionStatus_ArchivedSessionResumableWithMissingWorktree covers
+// the status half of the unarchive path: an archive-cancelled session whose
+// executors_running row survived cleanup must still report IsResumable, even
+// though its worktree directory is gone. Reporting "worktree not found" here
+// hid both the auto-resume and the manual resume button, so an unarchived task
+// had no way back at all.
+func TestGetTaskSessionStatus_ArchivedSessionResumableWithMissingWorktree(t *testing.T) {
+	ctx := context.Background()
+	repo := setupTestRepo(t)
+	seedTaskAndSession(t, repo, "task1", "session1", models.TaskSessionStateCancelled)
+
+	session, err := repo.GetTaskSession(ctx, "session1")
+	if err != nil {
+		t.Fatalf("failed to load session: %v", err)
+	}
+	session.AgentProfileID = "profile1"
+	session.ErrorMessage = models.SessionArchiveCancelReason
+	session.TaskEnvironmentID = "env1"
+	if err := repo.UpdateTaskSession(ctx, session); err != nil {
+		t.Fatalf("failed to update session: %v", err)
+	}
+	if err := repo.CreateTaskEnvironment(ctx, &models.TaskEnvironment{
+		ID: "env1", TaskID: "task1", ExecutorType: "worktree",
+		WorkspacePath: t.TempDir(), Status: models.TaskEnvironmentStatusReady,
+	}); err != nil {
+		t.Fatalf("CreateTaskEnvironment: %v", err)
+	}
+	// Recorded path the archive already deleted.
+	if err := repo.CreateTaskEnvironmentRepo(ctx, &models.TaskEnvironmentRepo{
+		ID:                "ter1",
+		TaskEnvironmentID: "env1",
+		WorktreeID:        "wid1",
+		RepositoryID:      "repo1",
+		WorktreePath:      filepath.Join(t.TempDir(), "archived-task", "kandev"),
+		WorktreeBranch:    "feature/archived-work",
+	}); err != nil {
+		t.Fatalf("CreateTaskEnvironmentRepo: %v", err)
+	}
+
+	now := time.Now().UTC()
+	if err := repo.UpsertExecutorRunning(ctx, &models.ExecutorRunning{
+		ID:          "er1",
+		SessionID:   "session1",
+		TaskID:      "task1",
+		Status:      "ready",
+		Resumable:   true,
+		ResumeToken: "acp-session-archived",
+		CreatedAt:   now,
+		UpdatedAt:   now,
+	}); err != nil {
+		t.Fatalf("failed to upsert executor running: %v", err)
+	}
+
+	taskRepo := newMockTaskRepo()
+	taskRepo.tasks["task1"] = &v1.Task{ID: "task1", State: v1.TaskStateInProgress}
+	agentMgr := &mockAgentManager{repoForExecutionLookup: repo}
+	svc := createTestServiceWithAgent(repo, newMockStepGetter(), taskRepo, agentMgr)
+	svc.executor = executor.NewExecutor(agentMgr, repo, testLogger(), executor.ExecutorConfig{})
+
+	resp, err := svc.GetTaskSessionStatus(ctx, "task1", "session1")
+	if err != nil {
+		t.Fatalf("GetTaskSessionStatus returned error: %v", err)
+	}
+	if resp.Error != "" {
+		t.Fatalf("expected no status error for a recreatable worktree, got %q", resp.Error)
+	}
+	if !resp.IsResumable {
+		t.Fatal("expected IsResumable=true: the worktree is recreated by the resume itself")
+	}
+	if !resp.NeedsResume {
+		t.Fatal("expected NeedsResume=true so the unarchived session auto-resumes")
+	}
+	if resp.ResumeReason != resumeReasonArchiveCancelledResumable {
+		t.Fatalf("expected ResumeReason=%q, got %q", resumeReasonArchiveCancelledResumable, resp.ResumeReason)
+	}
+}
@@ -1173,10 +1173,16 @@ func (m *Manager) recreate(ctx context.Context, existing *Worktree, req CreateRe
 		}
 	}

-	// Update record
+	// Update record. DeletedAt must be cleared alongside the status: archive
+	// cleanup releases the reference (status=deleted + deleted_at), and
+	// UpdateWorktree persists deleted_at verbatim — leaving it set would keep
+	// the freshly recreated worktree invisible to every lookup that filters on
+	// `deleted_at IS NULL`, so the next launch would mint a brand-new worktree
+	// beside the one we just restored.
 	now := time.Now()
 	existing.Path = worktreePath
 	existing.Status = StatusActive
+	existing.DeletedAt = nil
 	existing.UpdatedAt = now

 	if m.store != nil {
@@ -3,9 +3,12 @@ package worktree
 import (
 	"context"
 	"errors"
+	"os"
 	"path/filepath"
 	"strings"
 	"testing"
+
+	"github.com/kandev/kandev/internal/task/models"
 )

 // archiveDeletesLocalBranch simulates what task archive does to a worktree's
@@ -158,3 +161,76 @@ func TestRecreate_ForkPRFetchesPullHeadRef(t *testing.T) {
 		t.Errorf("worktree HEAD = %q, want %q (PR head)", gotSHA, prHeadSHA)
 	}
 }
+
+// TestCreate_RestoresReleasedWorktreeAfterArchive is the whole unarchive
+// round trip at the worktree layer. Archiving a task removes the worktree
+// directory and releases its reference (status=deleted + deleted_at) while
+// deliberately keeping the git branch, so the next launch must rebuild the
+// directory from the released record — including reactivating that record.
+// Leaving deleted_at set would hide the restored worktree from every lookup
+// that filters on `deleted_at IS NULL`, so the session would silently get a
+// brand-new worktree instead of its own work back.
+func TestCreate_RestoresReleasedWorktreeAfterArchive(t *testing.T) {
+	mgr, store := newReferenceCleanupTestManager(t)
+	ctx := context.Background()
+	seedReferenceCleanupSession(t, store, "task-archived", "session-archived", models.TaskSessionStateCompleted)
+
+	repoPath := initGitRepoWithRemote(t)
+	req := CreateRequest{
+		TaskID:         "task-archived",
+		SessionID:      "session-archived",
+		TaskTitle:      "Archived work",
+		RepositoryID:   "repository",
+		RepositoryPath: repoPath,
+		BaseBranch:     "main",
+		TaskDirName:    "task-archived",
+		RepoName:       "repository",
+	}
+	wt, err := mgr.Create(ctx, req)
+	if err != nil {
+		t.Fatalf("create worktree: %v", err)
+	}
+	// Work the user never pushed. It only survives because archive keeps the
+	// branch (DestroyWorktree passes removeBranch=false).
+	runGit(t, wt.Path, "commit", "--allow-empty", "-m", "unpushed work")
+	workSHA := strings.TrimSpace(runGit(t, wt.Path, "rev-parse", "HEAD"))
+
+	// Archive: remove the directory, release the reference, keep the branch.
+	if err := mgr.RemoveByID(ctx, wt.ID, false); err != nil {
+		t.Fatalf("archive worktree: %v", err)
+	}
+	if _, statErr := os.Stat(wt.Path); !os.IsNotExist(statErr) {
+		t.Fatalf("archive should remove the worktree directory, stat error = %v", statErr)
+	}
+
+	// Unarchive + resume: the launch carries the stored worktree ID.
+	resumeReq := req
+	resumeReq.WorktreeID = wt.ID
+	restored, err := mgr.Create(ctx, resumeReq)
+	if err != nil {
+		t.Fatalf("resume after unarchive must recreate the worktree: %v", err)
+	}
+	if restored.Path != wt.Path {
+		t.Fatalf("restored path = %q, want the original %q", restored.Path, wt.Path)
+	}
+	if restored.Branch != wt.Branch {
+		t.Fatalf("restored branch = %q, want the original %q", restored.Branch, wt.Branch)
+	}
+	if got := strings.TrimSpace(runGit(t, restored.Path, "rev-parse", "HEAD")); got != workSHA {
+		t.Fatalf("restored HEAD = %q, want the pre-archive work %q", got, workSHA)
+	}
+
+	// The record must be visible again to the session-scoped lookup the next
+	// launch uses; a lingering deleted_at would strand it.
+	found, err := store.GetWorktreeBySessionAndRepository(ctx, "session-archived", "repository", restored.BranchSlug)
+	if err != nil {
+		t.Fatalf("look up restored worktree: %v", err)
+	}
+	if found == nil {
+		t.Fatal("restored worktree is still hidden from the session lookup (deleted_at was not cleared)")
+	}
+	if found.ID != wt.ID {
+		t.Fatalf("session lookup returned worktree %q, want the restored %q", found.ID, wt.ID)
+	}
+	assertWorktreeReferenceStatus(t, store, wt.ID, StatusActive)
+}