fix: handle scenario where provisionerdserver deletes task before coderd (#21220)

DanielleMaywood · web-flow · commit c3224b793e96 · 2025-12-11T13:04:13.000Z
diff --git a/coderd/aitasks.go b/coderd/aitasks.go
@@ -2,6 +2,8 @@ package coderd
 
 import (
 	"context"
+	"database/sql"
+	"errors"
 	"fmt"
 	"net"
 	"net/http"
@@ -622,11 +624,15 @@ func (api *API) taskDelete(rw http.ResponseWriter, r *http.Request) {
 		}
 	}
 
+	// As an implementation detail of the workspace build transition, we also delete
+	// the associated task. This means that we have a race between provisionerdserver
+	// and here with deleting the task. In a real world scenario we'll never lose the
+	// race but we should still handle it anyways.
 	_, err := api.Database.DeleteTask(ctx, database.DeleteTaskParams{
 		ID:        task.ID,
 		DeletedAt: dbtime.Time(now),
 	})
-	if err != nil {
+	if err != nil && !errors.Is(err, sql.ErrNoRows) {
 		httpapi.Write(ctx, rw, http.StatusInternalServerError, codersdk.Response{
 			Message: "Failed to delete task",
 			Detail:  err.Error(),
diff --git a/coderd/provisionerdserver/provisionerdserver.go b/coderd/provisionerdserver/provisionerdserver.go
@@ -2267,6 +2267,13 @@ func (s *server) completeWorkspaceBuildJob(ctx context.Context, job database.Pro
 		if err != nil {
 			return xerrors.Errorf("update workspace deleted: %w", err)
 		}
+
+		// A user might delete their task workspace directly, instead of
+		// deleting the task. To avoid leaving the Task in a scenario where
+		// it has no workspace, we also attempt to delete the task.
+		//
+		// Deleting the task may fail if it has already been deleted as part
+		// of the typical task deletion workflow, so we explicitly allow that.
 		if workspace.TaskID.Valid {
 			if _, err := db.DeleteTask(ctx, database.DeleteTaskParams{
 				ID:        workspace.TaskID.UUID,
