Performance regression in EventsExecutor on jazzy

[jplapp]

Operating System:

ubuntu 22

ROS version or commit hash:

jazzy

RMW implementation (if applicable):

rmw_cyclonedds_cpp and rmw_fastdds_cpp

RMW Configuration (if applicable):

No response

Client library (if applicable):

rclcpp

'ros2 doctor --report' output

No response

Steps to reproduce issue

1. Run any action server with EventsExecutor, e.g. from ros2 demos , compile with EventsExecutor specified rclcpp_components_register_node(action_tutorials PLUGIN "action_tutorials_cpp::FibonacciActionServer" EXECUTABLE fibonacci_action_server EXECUTOR EventsExecutor)
2. ros2 run action_tutorials_cpp fibonacci_action_server
3. Check CPU load

Expected behavior

EventsExecutor has similar or less CPU load than SingleThreadedExecutor when running an idle action server

Running with SingleThreadedExecutor the CPU load is 0% (as nothing is happening)

Actual behavior

CPU load is significantly more than 0, e.g. 50%:

Additional information

When I revert #2674 , it behaves as expected

[MCFurry]

Ahh! We justed started a deep-dive and came to the same conclusion...

[ros-discourse]

This issue has been mentioned on ROS Discourse. There might be relevant details there:

https://discourse.ros.org/t/new-packages-and-patch-release-4-for-jazzy-jalisco-2025-04-07/43100/3

[mjcarroll]

@jmachowinski and @alsora FYI.

[jmachowinski]

I'll have a look

[jmachowinski]

@MCFurry @jplapp
Please check an verify that #2800 fixes the bug for you.

[MCFurry]

@jmachowinski I just ran a very quick test from your branch, but unfortunately the issue is still there... Thanks for the quick effort though!

[jmachowinski]

@jmachowinski I just ran a very quick test from your branch, but unfortunately the issue is still there... Thanks for the quick effort though!

How exactly did you test ? I checked with the procedure described in the first post, and the issue is solved for me using the patch. I just rechecked it on a second PC.

diff --git a/action_tutorials/action_tutorials_cpp/CMakeLists.txt b/action_tutorials/action_tutorials_cpp/CMakeLists.txt
index 56fa718..d3b1ea9 100644
--- a/action_tutorials/action_tutorials_cpp/CMakeLists.txt
+++ b/action_tutorials/action_tutorials_cpp/CMakeLists.txt
@@ -29,7 +29,7 @@ add_library(action_tutorials SHARED
   src/fibonacci_action_client.cpp
   src/fibonacci_action_server.cpp)
 rclcpp_components_register_node(action_tutorials PLUGIN "action_tutorials_cpp::FibonacciActionClient" EXECUTABLE fibonacci_action_client)
-rclcpp_components_register_node(action_tutorials PLUGIN "action_tutorials_cpp::FibonacciActionServer" EXECUTABLE fibonacci_action_server)
+rclcpp_components_register_node(action_tutorials PLUGIN "action_tutorials_cpp::FibonacciActionServer" EXECUTABLE fibonacci_action_server  EXECUTOR EventsExecutor)
 target_compile_definitions(action_tutorials
   PRIVATE "ACTION_TUTORIALS_CPP_BUILDING_DLL")
 ament_target_dependencies(action_tutorials

Without patch

With patch

[jplapp]

From my side I can verify that the issue is resolved, both in the test case as well as the full system.

@jmachowinski thanks for the quick fix!

[MCFurry]

Hmm my bad, it seems during my test only rclcpp was build from source but rclcpp_action was not. Now testing again and the issue indeed seems solved. Thanks again for the quick fix!