rmw_zenoh and ROS 2 daemon

[Yadunund]

With ROS 2, there is a long-running daemon in the background that stores information about the ROS graph to provide faster responses to queries, e.g. the list of node names, without having to discovery every other peer in the network. The daemon is automatically started when the relevant command-line tools are used for the first time.

With rmw_zenoh, we expect users to run the zenohd router which largely does the same above. However, there are issues with the zenohd router and daemon coexisting (mostly related to graph cache queries).

For example, if a user runs ros2 topic list without the zenohd router running, the process will check indefinitely for a router until one is started. Once the command exists, and if one kills the router, running ros2 topic list again will instantly return the topics from the daemon's cache. However in such scenarios we would like the cli tool to look for a router first before returning results from a cache.

Starting a discussion on how to manage such scenarios.

1. systemd service: An extention to the discussion in https://github.com/ros2/rmw_zenoh/issues/231, the service could also kill/restart the daemon as appropriate.
2. Disable daemon:
   • Add mechanisms in rmw API to selectively enable/disable certain "features" including the daemon.
   • Update rcl, rclpy and ros2cli such that these features are checked and only start the daemon if configured to do so.

[clalancette]

We had a bit of a discussion about this at the Waffle meeting today. Some notes:

• People were somewhat opposed to getting rid of the daemon, because it makes rmw_zenoh look "different" than other RMWs.
• One suggestion was to couple the lifetimes of the ROS 2 daemon and the Zenoh router. That would probably mean that when we launched the Zenoh router, we would also launch the ROS 2 daemon at the same time, and if one died they both die.
• Another suggestion was to just replace the ROS 2 daemon with the Zenoh router. There are a couple of major problems with that approach:
  • The protocol between the ROS 2 daemon the ros2cli tools is XMLRPC. So we'd have to have some kind of "wrapper" around the Zenoh router that converted XMLRPC calls into something that the Zenoh router understands.
  • The ROS 2 daemon doesn't always run. It is only started when certain ros2cli commands are invoked. In contrast, the Zenoh router is required and needs to be running at all times for rmw_zenoh to succeed, so it has different lifetime requirements.