Some replication messages might not be sent before next restart if the queue is full

When the queue of messages to sent is full (the maximum size is 10000 to not take too much memory), it will fail to add more to it and those might be waiting on the disk not being picked again before next restart.