Bug report

Required Info:

• Operating System:
  • Ubuntu bionic
• Installation type:
  • source
• Version or commit hash:
  • https://gist.github.com/sloretz/6e5d7324b115da574d79c7e785d8968e
• DDS implementation:
  • Fast-RTPS
• Client library (if applicable):
  • rclcpp

Steps to reproduce issue

1. Start top in a terminal
2. Start an action server using rmw_fastrtps_cpp

   RMW_IMPLEMENTATION=rmw_fastrtps_cpp ./build/test_communication/test_action_server_cpp Fibonacci /foo/bar
   

3. Start an action client using rmw_fastrtps_cpp

   RMW_IMPLEMENTATION=rmw_fastrtps_cpp ./build/test_communication/test_action_client_cpp Fibonacci /foo/bar
   

Expected behavior

I would expect negligible CPU usage after the goal has finished processing

Actual behavior

Server using 100% CPU after the goal has finished processing

Additional information

It does not happen using RMW connext.

If I debug the server, it seems that there is a loop inside rclcpp::spin() where rcl_wait() wakes indicating a timeout, the executor checks and nothing is ready so it goes back into rcl_wait() which wakes immediately with a timeout. The timeout given to rcl_wait() is -1 which becomes INT_MAX when passed to rmw_wait(). The 100% CPU usage ends when the action server expires the old goal (after 15 minutes by default).

Maybe there's an issue with how rcl_action_server is using the timer api (#343)?