I have this configuration:

Camera node -> Calibration Node

The calibration node is expected to sometimes crash (I know that this might not be the best design, however kind of stuck with this behaviour for now). To cover these chrashes we added the respawn=True parameter in the launch file with the hope that the system will just continue working. However this is not the case, the node becomes alive but no longer received images. We know this because in the calibration node we a a periodic callback that check how long it has been since an image was received, we see this contour going up meaning that no new images are received. In the camera node we log the publishing frequency and there we see that it keeps publishing at the expected frequency.

We are running ROS2 galactic in combination with Fast RTSP middleware with all nodes inside a docker container. Has anyone experienced this problem/found a solution?