I have a large production database in Amazon RDS MySQL (billions of rows) that I want to set up as a master, replicating to an Aurora Serverless instance. Unfortunately, it looks like the only way to copy the data over is by using mysqdump, which is going to be a serious pain in itself.
However, when I read about setting up a slave with this method, it looks like it requires the master to be basically locked down for the whole time that mysqdump is running, in order to preserve the binlog position, which makes it far too expensive in lost business.
I have already tried DMS - it appears to be unreliable, either losing data or, worse, inserting spurious rows, and I don't have the time or the insight to troubleshoot it. And as it turns out, snapshots cannot be restored into a serverless instance, for some reason.
Is there any way around these problem? I have seen mention of GTIDs, but so far I haven't found enough info to see if it is useful in this situation.