I've had a problem many times. When the task suddenly continues to back pressure, the back pressure node will no longer send any records unless the task is restarted. But I can confirm that it's not due to high pressure. During the back pressure period, the CPU utilization of the machine is all reduced, but not increased.
At present, I initially suspect that it has something to do with the Internet. Does anyone know the principle? Can network jitter cause this phenomenon?
