Martin Pecka

This is even documented: https://docs.ros.org/en/rolling/How-To-Guides/Working-with-multiple-RMW-implementations.html#ensuring-use-of-a-particular-rmw-implementation . But hell, can't this be handled better? There's even a solved issue here #6 that seems it solved this problem. But the problem is...

And what about adding the domain ID and RMW_IMPLEMENTATION to the answers of the daemon? That way, no new method would need to be introduced, and all clients could easily...

I agree that echo has the greatest value. Regarding bw/hz, I mostly imagine that you write a filtered subscription in your code and you want to get an idea about...

I've created an example repo where this can be tested: https://gitlab.fel.cvut.cz/peckama2/bug-geometry_tutorials . ```bash terminal1$ ros2 launch turtle_tf2_cpp turtle_tf2_demo.launch.py terminal2$ sleep 10 && ros2 component unload /turtle_container 1 ```

The stucking in single-threaded case is 100%. The segfault in multi-threaded case is not 100%, but I'd say 70%.

Okay, this is way worse than I thought. It happens not only with paused sim time, but in case of multi-threaded container, also with unpaused sim time and wall time....

Now I saw what I expected but haven't seen before (so probably a low-frequency thing): container segfaults when Ctrl+C-ing the whole launch file that runs both container and component. I...

I second this. I'd also like to have the option to add `iburst` to the server to get quicker convergence.

Ah, right, I see I can add `iburst` after the server address. But still it's missing the option to specify `server` instead of `pool`.

PR #144 should solve this.