Problem using rclpy.spin_until_future_complete inside Python plugin

[angel-ayala]

I'm using the Emitter send_string service call inside the step() python method, and when using rclpy.spin_until_future_complete() occurs a deadlock since the future never gets done. Apparently is because the step() never ends and therefore, the driver step never finish it iteration.

A workaround for this was using a condition to check Future.done() at every step() call but that leads to a information delay.

Can this be done differently?

[ygoumaz]

Are you using an asynchronous client call?

See https://docs.ros.org/en/humble/How-To-Guides/Sync-Vs-Async.html for more information. A solution using future.done() is also mentioned. I don't know any cleaner way to do it. I think I had tested the Emitter plugin when you submitted your PR and it was working fine. The delay shouldn't be over 1 timestep, or is it?

[angel-ayala]

The issue happen when call the Emitter.send_string C++ service inside the python plugin, and according how the driver works seems to be that way since the rclpp::spin_some() happens after the execution of the python step() method and needs to use the future.done() method to check for the service status.
However, this behavior only allow to call Emitter.send_string service in one step yes, and the next step no, why? I'm not sure. Maybe is related to the moment when rclpp::spin_some() and rclpy.spin_once() are called, requiring to be sync.

However it seems to be an issue to use C++ and Python ROS node in the same controller and how the Nodes spin call are synced. Should the ROS nodes be running in a separated thread just as the ros2_supervisor works?