If a transaction replay has to be executed twice due to a failure of the
original candidate master, the query queue could contain replayed
queries. The replayed queries would be placed into the queue if a new
connection needs to be created before the transaction replay can start.