Bug report

Obstacle_layer plugin cause high CPU usage in controller_Server and planner_server

Required Info:

• Operating System
  • Ubuntu 20.04
• ROS2 Version
  • Foxy Binaries
• Version or commit hash
  • foxy version and build From source
• DDS implementation
  • Fast-RTPS

Steps to reproduce issue

1. launch the laser node  and navigation2 software stack
2. keep the robot standstill or do task of moving point A to piont B
3. open htop in new terminal and you can find the cpu of controller_server and planner_server is unstable and high(sometimes the crest value may over 120%)

Expected behavior

The cpu usage of controller_server and planner_server is stable and not high

Actual behavior

The cpu usage of controller_server and planner_server is unstable and increase abruptly over 120%

Additional information

Temporary solution

1. Transform the scan data(laser_node) to pointcloud and then transform the pointcloud data from the laser_link to map before publish
2. Transform scan data from laser_link to map directly

The result of temporary solution

the cpu usage of controller_server and planner_server is decreased and keeping stable obviously(not increase abruptly)

Three points may cause this problem and Debug

1. projector_.transformLaserScanToPointCloud(message->header.frame_id, *message, cloud, *tf_) cause the most cpu in callback function -- through transform the scan to pointcloud data and not transform data from laser_link to map before publish, the bug is still alive
2. buffer->bufferCloud(cloud) -- ongoing debug
3. QOS buffer size is lager -- decrease the buffer size, the problem is still remaining

Co-woker

@tal-grossman @andersdu @duanyunxin

Although the temporary solution can fix this problem, we still could not find the point which cause this problem in obstacle_layer plugins. If anyone have ideas with this issue, we are hopefully Looking forward to your reply.
Thanks.