subscriber node uses 100% cpu when publisher disconnects (recent regression)

[rkent]

Generated by Generative AI

no

Operating System:

ubuntu 24

ROS version or commit hash:

rolling

RMW implementation (if applicable):

rmw_fastrtps_cpp

RMW Configuration (if applicable):

No response

Client library (if applicable):

rclcpp

'ros2 doctor --report' output

ros2 doctor --report

<COPY OUTPUT HERE>

Steps to reproduce issue

1. Install a recent version of rolling (sudo apt update & sudo apt upgrade)
2. Install ros-rolling-demo-nodes-cpp
3. On one terminal, ros2 run demo_nodes_cpp listener
4. On another terminal, run ros2 run demo_nodes_cpp talker then after a few messages, kill with ^c

Expected behavior

Very low cpu usage on the listener

Actual behavior

100% cpu usage on the listener, as viewed in top

Additional information

I initially discovered this in docker, so I ran a few docker images. This does not occur in humble or kilted docker images.

Most interesting is rolling. osrf/ros:rolling-desktop does not show the bug. However, if you run apt update && apt upgrade there, then the bug DOES appear. So this is a fairly recent (and IMHO serious) regression.

I'm not sure it is occurring in rclcpp of course. profiling showed fast-ddr was involved in the behavior, as well of course as all of the intermediate rcl and rmw.

[jmachowinski]

Is this only happening with fastdds or with cyclone as well ?

[mjcarroll]

I don't see anything obvious from the changelog for rclcpp, but as you said, there are a few levels where this could be happening.

[clalancette]

Is this only happening with fastdds or with cyclone as well ?

Running with RWM_IMPLEMENTATION=rmw_cyclonedds_cpp or RMW_IMPLEMENTATION=rmw_zenoh_cpp doesn't reproduce the problem for me locally, while RMW_IMPLEMENTATION=rmw_fastrtps_cpp does.

[mjcarroll]

Is this only happening with fastdds or with cyclone as well ?

Running with RWM_IMPLEMENTATION=rmw_cyclonedds_cpp or RMW_IMPLEMENTATION=rmw_zenoh_cpp doesn't reproduce the problem for me locally, while RMW_IMPLEMENTATION=rmw_fastrtps_cpp does.

Interesting, rmw_fastrtps has no changes lately. @MiguelCompany have you done a release of the underlying middleware recently?

[rkent]

Here's a regression range.

https://ci.ros2.org/view/packaging/job/packaging_linux/4041/ (Dec 16, 2025) does not have the bug.

https://ci.ros2.org/view/packaging/job/packaging_linux/4042/ (Dec 17, 2025) has the bug.

[rsanchez15]

Hi @rkent

We've been able to reproduce the issue on our end and are working to fix it. The fix will be available in a patch version of Fast DDS and also ROS 2 Rolling very soon.

[MiguelCompany]

@rkent eProsima/Fast-DDS#6276 should fix this

[ros-discourse]

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

https://discourse.openrobotics.org/t/irobots-ros-benchmarking-suite-now-available/51921/5

[MiguelCompany]

@rkent We just released Fast DDS 3.4.2, which includes the fix for this.
This means that the issue is fixed when building from sources, and that the next rolling sync would fix binaries when doing apt upgrade