ROS2 communication stopped between nodes on same machine when network is down

[robert-preissl]

Bug report

Required Info:

• Operating System:
  • Ubuntu 22.04.03
• Installation type:
  • binaries (running the latest ros2/iron docker image)
• Version or commit hash:
  • ros2 Iron
• DDS implementation:
  • ros-iron-cyclonedds 0.10.4-1jammy
  • ros-iron-rmw-cyclonedds-cpp 1.6.0-2jammy
• Client library (if applicable):
  • rclcpp

Steps to reproduce issue

The setup is as follow: using one computer, connected to a wifi network, two terminals both running a ros2 iron docker container (with --network=host)
(the wireless interface as well as the loopback lo interface are configured as Multicast)

• run in one terminal the ros2 listener node (environment variables are the RMW_IMPLEMENTATION= rmw_cyclonedds_cpp):
  ros2 run demo_nodes_cpp listener

• in a second terminal run the ros2 talker node (same environment variables as above):
  ros2 run demo_nodes_cpp talker

• the terminal with the listener prints Hello World messages with increasing ids. (1,2,3, etc.)

• Now, we disconnect the wifi on the computer

• the terminal with the listener does not print new Hello world messages

• the terminal of the talker prints error messages like tev: ddsi_upd_conn_write to udp/192.168.128.127:21163 failed with retcode -3

Expected behavior

• since all the communication is local (i.e., on the same machine), an interruption in the wifi network should not interrupt local ROS communication

Actual behavior

• communication from the talker to the listener (all on the same host -- but different docker containers) is impacted.

[wjwwood]

@eboasson may be able to comment here

[eboasson]

I do have some ideas, but I do need to do a bit of checking to be sure what exactly is causing it, and what would be the solution (workaround, change to Cyclone, whatever). I did want to let you know I am now—belatedly—aware of the issue.

[robert-preissl]

Thank you @eboasson