Just to clarify, our application does not close and reopen the fd without restarting the container, so we don’t know if that would resume the data. We do know that restarting the container (and therefore closing the fd) without restarting the device does resolve the issue.
I’m running a test now just doing a cat inside the container (I’m catting to a file and watching the file size since I didn’t want to risk the console printing of the high rate binary data affecting the test). Hopefully that fails eventually and this isn’t something we can only replicate via the fd open inside of our application. When it fails I can check for other processes and then try killing it and starting again without restarting the container. After that I’ll try the host OS. Unfortunately it takes a random amount of time to fail so I don’t know when it’ll happen, or even if it will under these conditions, but I’ll let you know once it does.