In the graph you can clearly see, the 2 stop upgrade where the DB is failed over to the backup AZ and then upgraded before being failed back again. Now that the instance has 8 vCPU its only exceeded that once during the backups.