We have identified the root cause of the problem. One of the instances was experiencing latency issues. This resulted in a reset of that instance by our cloud platform provider. We were already aware of this risk. The solution is already under construction by our development team.