We use AWS DMS to migrate (initial load + cdc) AWS Aurora MySQL data into our Redshift (in the future we are going to use ICEBERG). In case there is any issue with DMS (RI break, Task break, MySQL Major upgrade) – we recreating or restarting the task.
This force us to code in a way where we expect failures to happen, so we check at the beginning of our scripts if we are in sync, otherwise wait. Also, since there are so many updates, long SELECTs are failing on deadlocks which forces us to clone the high frequency tables.