AWS DMS continuous replication latency question

Question

I am working on a project to migrate service from one database instance to another. My intention is to use AWS DMS to do replication from an RDS MySQL (source) instance to another RDS MySQL (target) instance and then cut the service over from connecting from the "source" to the "target" DB. I plan on having a task with Full load + CDC and doing the cutover only when on the CDC portion of the task

I am trying to plan on how much downtime I am going to need to take and I'm curious what kind of latency there is between writing to the source and replication to the target. Is it on the magnitude of ms, seconds, or minutes?

Answer 1

Usually milliseconds, but this depends on the type of update you are executing.

For example, suppose you do an UPDATE that takes 5 minutes to execute on the primary instance. It isn't logged until it finishes and the transaction commits. Then there's some latency to transfer the log via CDC to the replica. The replica reads it, saves it to its own local log, and then starts executing it.

This means the 5 minute update doesn't start on the replica until it has finished on the primary and then there's some extra latency in between for the logging and the transfer. So the best case is that it will take 5 more minutes.

Of course most updates you would run are individually a lot quicker than 5 minutes. That's why I say it depends on your updates. There's no way AWS can predict your downtime, nor can we.