From a recent run we noticed that the client-replica connection died at some point but neither the client nor the replica noticed this, resulting the client logging empty responses.
Theoretically this could happen between replicas too, which is perfectly fine in a production settings as network partitions are common, but we need to instrument both the client and the replica to restart the connections after some timeouts.
From a recent run we noticed that the client-replica connection died at some point but neither the client nor the replica noticed this, resulting the client logging empty responses.
Theoretically this could happen between replicas too, which is perfectly fine in a production settings as network partitions are common, but we need to instrument both the client and the replica to restart the connections after some timeouts.