fix(coderd/provisionerdserver): workaround lack of coder_ai_task resource on stop transition (coder#19560)

johnstcn · mafredri · web-flow · commit bd139f3a436c · 2025-08-27T10:33:17.000+01:00
This works around the issue where a task may "disappear" on stop.
Re-using the previous value of `has_ai_task` and `sidebar_app_id` on a
stop transition.

---------

Co-authored-by: Mathias Fredriksson &lt;mafredri@gmail.com&gt;
diff --git a/coderd/provisionerdserver/provisionerdserver.go b/coderd/provisionerdserver/provisionerdserver.go
@@ -1995,6 +1995,37 @@ func (s *server) completeWorkspaceBuildJob(ctx context.Context, job database.Pro
 			sidebarAppID = uuid.NullUUID{UUID: id, Valid: true}
 		}
 
+		// This is a hacky workaround for the issue with tasks 'disappearing' on stop:
+		// reuse has_ai_task and sidebar_app_id from the previous build.
+		// This workaround should be removed as soon as possible.
+		if workspaceBuild.Transition == database.WorkspaceTransitionStop && workspaceBuild.BuildNumber > 1 {
+			if prevBuild, err := s.Database.GetWorkspaceBuildByWorkspaceIDAndBuildNumber(ctx, database.GetWorkspaceBuildByWorkspaceIDAndBuildNumberParams{
+				WorkspaceID: workspaceBuild.WorkspaceID,
+				BuildNumber: workspaceBuild.BuildNumber - 1,
+			}); err == nil {
+				hasAITask = prevBuild.HasAITask.Bool
+				sidebarAppID = prevBuild.AITaskSidebarAppID
+				warnUnknownSidebarAppID = false
+				s.Logger.Debug(ctx, "task workaround: reused has_ai_task and sidebar_app_id from previous build to keep track of task",
+					slog.F("job_id", job.ID.String()),
+					slog.F("build_number", prevBuild.BuildNumber),
+					slog.F("workspace_id", workspace.ID),
+					slog.F("workspace_build_id", workspaceBuild.ID),
+					slog.F("transition", string(workspaceBuild.Transition)),
+					slog.F("sidebar_app_id", sidebarAppID.UUID),
+					slog.F("has_ai_task", hasAITask),
+				)
+			} else {
+				s.Logger.Error(ctx, "task workaround: tracking via has_ai_task and sidebar_app from previous build failed",
+					slog.Error(err),
+					slog.F("job_id", job.ID.String()),
+					slog.F("workspace_id", workspace.ID),
+					slog.F("workspace_build_id", workspaceBuild.ID),
+					slog.F("transition", string(workspaceBuild.Transition)),
+				)
+			}
+		}
+
 		if warnUnknownSidebarAppID {
 			// Ref: https://github.com/coder/coder/issues/18776
 			// This can happen for a number of reasons:
diff --git a/coderd/provisionerdserver/provisionerdserver_test.go b/coderd/provisionerdserver/provisionerdserver_test.go
@@ -2842,9 +2842,12 @@ func TestCompleteJob(t *testing.T) {
 		// has_ai_task has a default value of nil, but once the workspace build completes it will have a value;
 		// it is set to "true" if the related template has any coder_ai_task resources defined, and its sidebar app ID
 		// will be set as well in that case.
+		// HACK(johnstcn): we also set it to "true" if any _previous_ workspace builds ever had it set to "true".
+		// This is to avoid tasks "disappearing" when you stop them.
 		t.Run("WorkspaceBuild", func(t *testing.T) {
 			type testcase struct {
 				name             string
+				seedFunc         func(context.Context, testing.TB, database.Store) error // If you need to insert other resources
 				transition       database.WorkspaceTransition
 				input            *proto.CompletedJob_WorkspaceBuild
 				expectHasAiTask  bool
@@ -2944,6 +2947,17 @@ func TestCompleteJob(t *testing.T) {
 					expectHasAiTask:  true,
 					expectUsageEvent: false,
 				},
+				{
+					name:       "current build does not have ai task but previous build did",
+					seedFunc:   seedPreviousWorkspaceStartWithAITask,
+					transition: database.WorkspaceTransitionStop,
+					input: &proto.CompletedJob_WorkspaceBuild{
+						AiTasks:   []*sdkproto.AITask{},
+						Resources: []*sdkproto.Resource{},
+					},
+					expectHasAiTask:  true,
+					expectUsageEvent: false,
+				},
 			} {
 				t.Run(tc.name, func(t *testing.T) {
 					t.Parallel()
@@ -2980,6 +2994,9 @@ func TestCompleteJob(t *testing.T) {
 					})
 
 					ctx := testutil.Context(t, testutil.WaitShort)
+					if tc.seedFunc != nil {
+						require.NoError(t, tc.seedFunc(ctx, t, db))
+					}
 
 					buildJobID := uuid.New()
 					wsBuildID := uuid.New()
@@ -2999,8 +3016,13 @@ func TestCompleteJob(t *testing.T) {
 						Tags:          pd.Tags,
 					})
 					require.NoError(t, err)
+					var buildNum int32
+					if latestBuild, err := db.GetLatestWorkspaceBuildByWorkspaceID(ctx, workspaceTable.ID); err == nil {
+						buildNum = latestBuild.BuildNumber
+					}
 					build := dbgen.WorkspaceBuild(t, db, database.WorkspaceBuild{
 						ID:                wsBuildID,
+						BuildNumber:       buildNum + 1,
 						JobID:             buildJobID,
 						WorkspaceID:       workspaceTable.ID,
 						TemplateVersionID: version.ID,
@@ -3038,7 +3060,7 @@ func TestCompleteJob(t *testing.T) {
 					require.True(t, build.HasAITask.Valid) // We ALWAYS expect a value to be set, therefore not nil, i.e. valid = true.
 					require.Equal(t, tc.expectHasAiTask, build.HasAITask.Bool)
 
-					if tc.expectHasAiTask {
+					if tc.expectHasAiTask && build.Transition != database.WorkspaceTransitionStop {
 						require.Equal(t, sidebarAppID, build.AITaskSidebarAppID.UUID.String())
 					}
 
@@ -4244,3 +4266,63 @@ func (f *fakeUsageInserter) InsertDiscreteUsageEvent(_ context.Context, _ databa
 	f.collectedEvents = append(f.collectedEvents, event)
 	return nil
 }
+
+func seedPreviousWorkspaceStartWithAITask(ctx context.Context, t testing.TB, db database.Store) error {
+	t.Helper()
+	// If the below looks slightly convoluted, that's because it is.
+	// The workspace doesn't yet have a latest build, so querying all
+	// workspaces will fail.
+	tpls, err := db.GetTemplates(ctx)
+	if err != nil {
+		return xerrors.Errorf("seedFunc: get template: %w", err)
+	}
+	if len(tpls) != 1 {
+		return xerrors.Errorf("seedFunc: expected exactly one template, got %d", len(tpls))
+	}
+	ws, err := db.GetWorkspacesByTemplateID(ctx, tpls[0].ID)
+	if err != nil {
+		return xerrors.Errorf("seedFunc: get workspaces: %w", err)
+	}
+	if len(ws) != 1 {
+		return xerrors.Errorf("seedFunc: expected exactly one workspace, got %d", len(ws))
+	}
+	w := ws[0]
+	prevJob := dbgen.ProvisionerJob(t, db, nil, database.ProvisionerJob{
+		OrganizationID: w.OrganizationID,
+		InitiatorID:    w.OwnerID,
+		Type:           database.ProvisionerJobTypeWorkspaceBuild,
+	})
+	tvs, err := db.GetTemplateVersionsByTemplateID(ctx, database.GetTemplateVersionsByTemplateIDParams{
+		TemplateID: tpls[0].ID,
+	})
+	if err != nil {
+		return xerrors.Errorf("seedFunc: get template version: %w", err)
+	}
+	if len(tvs) != 1 {
+		return xerrors.Errorf("seedFunc: expected exactly one template version, got %d", len(tvs))
+	}
+	if tpls[0].ActiveVersionID == uuid.Nil {
+		return xerrors.Errorf("seedFunc: active version id is nil")
+	}
+	res := dbgen.WorkspaceResource(t, db, database.WorkspaceResource{
+		JobID: prevJob.ID,
+	})
+	agt := dbgen.WorkspaceAgent(t, db, database.WorkspaceAgent{
+		ResourceID: res.ID,
+	})
+	wa := dbgen.WorkspaceApp(t, db, database.WorkspaceApp{
+		AgentID: agt.ID,
+	})
+	_ = dbgen.WorkspaceBuild(t, db, database.WorkspaceBuild{
+		BuildNumber:        1,
+		HasAITask:          sql.NullBool{Valid: true, Bool: true},
+		AITaskSidebarAppID: uuid.NullUUID{Valid: true, UUID: wa.ID},
+		ID:                 w.ID,
+		InitiatorID:        w.OwnerID,
+		JobID:              prevJob.ID,
+		TemplateVersionID:  tvs[0].ID,
+		Transition:         database.WorkspaceTransitionStart,
+		WorkspaceID:        w.ID,
+	})
+	return nil
+}
