Memory leak slowly and chunk size increase with multiple nodes

[Lannister-Xiaolin]

Bug report

Required Info:

• Operating System: Ubuntu20.04 Arm64 Galactic
• Installation type:
• binaries
• DDS implementation:

  • cyclonedds

• Client library (if applicable):

  • rclcpp

Steps to reproduce issue

There there nodes, one is for image node witch publish fixed size image by zero copy mode Then there two nodes(name: NodeA and Node B) subscribe to the fixed image, and publish some point cloud data(without zero copy mode) For these nodes, we set env variables RMW_IMPLEMENTATION and CYCLONEDDS_URI correctly Then, when I echo point cloud published by (NodeA and Node B) in the terminal without setting RMW_IMPLEMENTATION and CYCLONEDDS_URI. We can see memoy increase

If CPU cost is very hight and callback processing message take two much time, chunk usage and memory cost for nodes will increase immediatly, not never come down after cpu usage come down.

Expected behavior

Memory usage and chunk usage keep stable

Actual behavior

Memory and chunk usage keep increasing

[clalancette]

Which ROS 2 release are you using?

[Lannister-Xiaolin]

Which ROS 2 release are you using?

Galactic

[clalancette]

Galactic

Can you try with Humble and see if the situation is any better? There were a few upgrades to the iceoryx integration in that release. Note that if you want to stick with Ubuntu 20.04, you'll have to build Humble from source: http://docs.ros.org/en/humble/Installation/Alternatives/Ubuntu-Development-Setup.html

[tdinesh]

365 solves this for rolling.

For galactic I have a backport at galactic 365