No Documentation on simulation time behavior of ros2_control

[FranekStark]

I am trying to figure out the behavior of ROS 2 Control running with a simulation, which does not run on real-time.
We want to use ROS 2 Control for a very complex system where (our custom) simulation might not be able to run in real-time.

I guess my question splits up into two directions, depending on how the simulation is 'connected' to ros2-control:

1. The simulation is running within a HW component:
   I assume one way of integrating a simulation into ROS 2 Control is to run it within a HardwareComponent. Is that the case for example for the gz_ros2_control.
   My question would now be, where at best I run the "actual" simulation/integration (which blocks and might take longer than the controller period), in the read() or in the write() callback? I would say I would best run the simulation in the write() with the controller's period as the time to simulate. How would ROS 2 Control react to the blocking call and the difference in simulation and system time.

2. The simulation is connected via topics like in [topic_based_ros2_control]:(https://github.com/PickNikRobotics/topic_based_ros2_control/tree/main)
   As in our case the simulation is computationally very heavy it makes sens to use this approach, as it allows the simulation to run on a different machine.
   Here, as in topic_based_ros2_control, during read() I would use the latest states received via subscriber and in write() I would asynchronously publish the commands.
   The simulation itself publishes the simulated time on the /clock topic. How it the behavior of ROS 2 Control, when it is launched with --use-sim-time. Is it following the /clock no matter what, or are there any things I have to note?

Unfortunately, I could not find any documentation on this important behaviors, and there are quite some issues/PR's on that matter (https://github.com/ros-controls/ros2_control/issues?q=simulation+clock).

Thanks for any help or advise. - After I have figured all of that out, I would be very happy to provide a best-practice or section for the documentation.

[christophfroehlich]

Hi!
about gz_ros2_control: The read/udpate call is done here triggered by the physics engine update. There is no clock involved in the controller_manager then, it just depends on the arguments of the CM calls. (no ros2_control_node involved)

For the second point, the ros2_control_node should handle this now
#859
#1810