how to gracefully terminate a python node

[christophfroehlich]

I'm a little bit puzzled about the fact that the code in the finally branch has zero coverage. maybe rlcpy.ok() is always false here. Should the destroy_node() be called even if the context is not valid any more? @saikishor any idea?

Originally posted by @christophfroehlich in #1369 (comment)

This does not happen always, but if it enters there is a different error

2024-11-22T12:11:30.3702266Z [publisher_joint_trajectory_controller-2] Keyboard interrupt received. Shutting down node.
2024-11-22T12:11:30.3703121Z [publisher_joint_trajectory_controller-2] Traceback (most recent call last):
2024-11-22T12:11:30.3704880Z [publisher_joint_trajectory_controller-2]   File "/home/runner/work/ros2_controllers/ros2_controllers/.work/target_ws/install/ros2_controllers_test_nodes/lib/ros2_controllers_test_nodes/publisher_joint_trajectory_controller", line 33, in <module>
2024-11-22T12:11:30.3706968Z [publisher_joint_trajectory_controller-2]     sys.exit(load_entry_point('ros2-controllers-test-nodes==4.16.0', 'console_scripts', 'publisher_joint_trajectory_controller')())
2024-11-22T12:11:30.3708246Z [publisher_joint_trajectory_controller-2]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-22T12:11:30.3710170Z [publisher_joint_trajectory_controller-2]   File "/home/runner/work/ros2_controllers/ros2_controllers/.work/target_ws/install/ros2_controllers_test_nodes/lib/python3.12/site-packages/ros2_controllers_test_nodes/publisher_joint_trajectory_controller.py", line 196, in main
2024-11-22T12:11:30.3711758Z [publisher_joint_trajectory_controller-2]     rclpy.shutdown()
2024-11-22T12:11:30.3712778Z [publisher_joint_trajectory_controller-2]   File "/opt/ros/rolling/lib/python3.12/site-packages/rclpy/__init__.py", line 170, in shutdown
2024-11-22T12:11:30.3713790Z [publisher_joint_trajectory_controller-2]     _shutdown(context=context)
2024-11-22T12:11:30.9113697Z [publisher_joint_trajectory_controller-2]   File "/opt/ros/rolling/lib/python3.12/site-packages/rclpy/utilities.py", line 82, in shutdown
2024-11-22T12:11:30.9115644Z [publisher_joint_trajectory_controller-2]     context.shutdown()
2024-11-22T12:11:30.9117325Z [publisher_joint_trajectory_controller-2]   File "/opt/ros/rolling/lib/python3.12/site-packages/rclpy/context.py", line 179, in shutdown
2024-11-22T12:11:30.9118481Z [publisher_joint_trajectory_controller-2]     self._cleanup()
2024-11-22T12:11:30.9119840Z [publisher_joint_trajectory_controller-2]   File "/opt/ros/rolling/lib/python3.12/site-packages/rclpy/context.py", line 170, in _cleanup
2024-11-22T12:11:30.9121292Z [publisher_joint_trajectory_controller-2]     self.__context.shutdown()
2024-11-22T12:11:30.9122665Z [publisher_joint_trajectory_controller-2] rclpy._rclpy_pybind11.RCLError: failed to shutdown: rcl_shutdown already called on the given context, at ./src/rcl/init.c:290
2024-11-22T12:11:30.9125963Z [ERROR] [publisher_joint_trajectory_controller-2]: process has died [pid 30093, exit code 1, cmd '/home/runner/work/ros2_controllers/ros2_controllers/.work/target_ws/install/ros2_controllers_test_nodes/lib/ros2_controllers_test_nodes/publisher_joint_trajectory_controller --ros-args --params-file /home/runner/work/ros2_controllers/ros2_controllers/.work/target_ws/install/ros2_controllers_test_nodes/share/ros2_controllers_test_nodes/test/

https://github.com/ros-controls/ros2_controllers/actions/runs/11972094365/job/33378268325?pr=1314

[saikishor]

Thanks for creating an issue!

[firesurfer]

I did some research when looking into this: ros2/rclpy#1287 and if I understood it correctly
the context becomes invalid after leaving the spin with an exception.

In our main python node we have this weird looking construct:

    retry_counter = 0
    while rclpy.ok():
        try:
            rclpy.spin(node)
            retry_counter += 1
        except (KeyboardInterrupt, rclpy.executors.ExternalShutdownException):
            break
        except Exception as ex:
            node.get_logger().error(str(ex))
            node.get_logger().error(str(traceback.format_exc()))

            if retry_counter > 10:
                break

The retry_counter was/(is ?) a workaround for a just once in a while occurring issue with rclpy where it would throw an exception but the context was still okey. The loggers are still operational afterwards btw. but do not log to ros_out anymore - only to the stdout.

TLDR:
I would say: Do not call shutdown when exiting spin with an exception as the context is already destroyed.
If you need ROS communication afterwards: You can create a new context and create new node in that context.