Bug report

Required Info:

• Operating System:
  Ubuntu 20.04 (amd64 & arm64)

• Installation type:
  binaries

• Version or commit hash:

• DDS implementation:
  Fast-RTPS & Cyclon

• Client library (if applicable):
  rclcpp

Steps to reproduce issue

The degradation is reproduced mostly but not only when running the nav2 both on physical robot and with the tb3 simulation.

• The fastest way is to run the nav2 and manually or with Behavior tree to frequently send_goal to an action server

• CTRL + C get always this msg:
  [waypoint_follower-5] [1608026728.932168403][rclcpp_action][ERROR]: Error in destruction of rcl action client handle: the Node Handle was destructed too early. You will leak memory

Expected behavior

• send_goal will be successful when asked repeatedly in medium-high frequencies
• no memory leak from nodes
• general stability improvement

Actual behavior

It seems actions and topics sometimes highly unstable, for example:

• frequently action server get this error msg:
  [bt_navigator-4] [1608026552.800316163][bt_navigator][ERROR]: Action server failed while executing action callback: "send_goal failed"

• after that happens sometimes nodes (even Rviz) stop getting msgs and stop receiving TFs - it requires restart of the entire ros system when it happens

• stability in general feels degraded and I require to restart the processes many times

Additional information

Talking to @SteveMacenski - Other users experienced similar behaviors on the ros2 master as well.