That was the mind-boggling part to me. After testing failover and noticing the behavior we had a number of discussions with AWS reps and support engineers that confirmed it. And Aurora's failover documentation hints at it but downplays the severity.