xqms
commented
3 years ago Yes that would be great indeed. Since I'm not using ROS2 (yet), I don't have any motivation to work on it, though. Here are some thoughts that might be relevant to someone looking into this:
- The rosmon project started in a pretty special situation: My team for the DARPA Robotics Challenge needed a way to supervise large amounts of nodes, restart nodes dynamically, and be sure the monitoring system would not crash. Since no one of us was very proficient in Python (and I still think it's a bad idea to write mission-critical stuff in dynamic languages), this led to a complete rewrite of the roslaunch logic in C++. While this had unexpected benefits (I learned a lot about the hidden intricacies and bugs in the roslaunch XML parser), I think that bolting a python-based ncurses UI onto roslaunch would also have been a solution, and this may very well be a better/easier solution for ROS2.
- It seems that the launch system in ROS2 is even more python-based than in ROS1 - capable of generating node constellations dynamically. I have no idea how well an alternative launcher like rosmon can be adapted to work with it. If the python part can be used in a way that spits out command lines for processes and rosmon can then take over the monitoring part, it could work.
- Apart from the launch file parser, which would need to be adapted to whatever ROS2 uses, the ROS interface (topics/services) of rosmon would also need to be ported.
- Because of the tight integration with roscpp, most parts of rosmon would need to be touched for a port of ROS2. So it might be better to fork (or completely rewrite) rosmon for ROS2.
I'm going to leave this issue open to collect information and interested users as well as contributors. As stated above, I won't invest my time into ROS2 (yet).
buckleytoby
gibsonhu123
It would be great to if rosmon could work with ROS2.