ReplSetConnection raises except when a secondary node restarts, even though read secondary is set to false.

In our case, we have 5 MongoDB nodes in a ReplSet. Restart a non-primary node and the client will likely shit itself, mid operation for no explicable reason.

Whatever is going on should be caught and a retry would be ideal.

Increasing the size of a ReplicaSet seems to increase the likelihood of failure for clients. So, more Mongoz = more risk. Except that less Mongoz = more risk.