Occasional crash with SJTC

[Schnilz]

Hi, first of all thanks for writing this, it is awesome :) The driver works realy well and setting it up was pretty straight forward because of the nice documentation, but i think i encountered a bug. Most of the time this issue doesn't occur but sometimes the SJTC crashes with following log messages:

[ur_ros2_control_node-1] [INFO] [1652881749.579905572] [scaled_joint_trajectory_controller]: Received new action goal
[ur_ros2_control_node-1] [INFO] [1652881749.580010468] [scaled_joint_trajectory_controller]: Accepted new action goal
[ur_ros2_control_node-1] terminate called after throwing an instance of 'std::runtime_error'
[ur_ros2_control_node-1]   what():  Executing action server but nothing is ready
[ERROR] [ur_ros2_control_node-1]: process has died [pid 57695, exit code -6, cmd ...

the error comes from rclcpp_action/src/server.cpp line 306 but im not realy sure what it actualy means :/ Like i said, reproducing it is not so easy as it works most of the time. I have a feeling it happens more often when the trajectory is short, but this could be wrong....

[fmauch]

Thanks for posting this. This is definitively something to look into. Could you tell us a little about your setup for us to reproduce this?

• ROS version
• This repo version (commit hash ideally)

[Schnilz]

The newest commit on the glactic banch (so ae85525) with ros galactic. It is maybe caused by me not using a realtime kernel on my dual-core thinkpad which run the driver. I think that because if you put a delay (like printing to cout 10000 times) at the end of the JTC feedback callback setup function the error basicly never happens anymore (hit it once in 2h of running, compared to every ~2min before).

[fmauch]

This issue is not forgotten, but will need some more time to reproduce.

[RoyCLR]

I seems to hit the same issue, but with joint_trajectory_controller:

[ros2_control_node-1] [1661228175.046757625INFO] [joint_trajectory_controller::goal_callback]: Received new action goal
[ros2_control_node-1] [1661228175.046794670INFO] [joint_trajectory_controller::goal_callback]: Accepted new action goal
[ros2_control_node-1] terminate called after throwing an instance of 'std::runtime_error'
[ros2_control_node-1]   what():  Executing action server but nothing is ready
[ERROR] [ros2_control_node-1]: process has died [pid 1708770, exit code -6, cmd

Setup:

• Using ROS2 Foxy
• Universal_Robots_ROS2_Driver at 69c2030fe88861039826f838aeca0ffc03b5a303
• With close to, but non-real time kernel "Liquorix" https://liquorix.net/

[fmauch]

Coming back to this: https://github.com/ros2/rclcpp/pull/2250

I've had this on my local long-term test and the linked PR fixes that for me.

[tommy-erko]

Detailed explanation of the issue is here: https://github.com/ros2/rclcpp/issues/2242

[firesurfer]

@fmauch Did you figure out if that fix is also backported to iron or humble ? Apparently it is finally merged into rolling but I couldn't figure out if there are plans to backport it

[fmauch]

https://github.com/ros2/rclcpp/issues/2242#issuecomment-2072845701 not necessarily :-/