Do what I said. It will stop.
You may have your cluster compromised due some corruption. It may the timestamps, the data itself, or the data in the disk have a corruption due some disk failure. If you have HA this wont happen.
If you give low resources to your cluster, mean you don’t give a correct dimensioning to your cluster, it will end up like that. That best way to recover from that is exporting and doing it right.