ROS2: High CPU load of python scripts
Closed this issue · 10 comments
Getting the following error:
[ERROR] [ensenso_camera_node-10]: process has died [pid 17673, exit code -9
when running the ensenso_camera_node with a virtual camera and virtual objects. It crashes after around ~300-350 seconds.
Is looks like it is a memory problem. The memory of the ensenso_camera_node always increases until it crashes with the error code. The virtual camera also disappears from NxView after the crash, and I have to restart machine to get it back.
Investigating further I see that I have another node running of type image_stream in the ensenso_camera package. This node's CPU usage increases (up to 100 % 😲 ) as the nodes run.
Any ideas what might cause this @benthie? I believe it has to do with the requested data from the image_stream node...
I will have a look at it.
- Are you using ROS1 or ROS2?
- Are you using the virtual camera directly or as a file camera?
Let me summarize what you are doing: you have an ensenso_camera_node running and besides that only the image_stream script?
I will have a look at it.
* Are you using ROS1 or ROS2?
ROS2
* Are you using the virtual camera directly or as a file camera?
Virtual camera directly
Let me summarize what you are doing: you have an
ensenso_camera_noderunning and besides that only theimage_streamscript?
Yes, I start them both from a launch file. I also have some other nodes running TF, robot state publishers, but I do not think they cause any problems here.
Thank you for the details. I will get back to you if I have found something .
Which Ensenso SDK version do you use?
Which Ensenso SDK version do you use?
3.3.1417
I think I found a solution for the high CPU load of the python scripts. The fix is to use a MultiThreadedExecutor in combination with a ReentrantCallbackGroup. I am currently testing it on my machine and it looks good with regard to the CPU usage. For the memory leak I created a separate issue. I will keep you posted as soon as I have a working fix that also passes all tests.