How Does Migrating to Aurora Global Database Affect Performance?

We've been using the Aurora Global Database with MySQL for nearly a year. The writer performance remained stable, and the storage-level replication works as promised. However, during heavy write operations like bulk imports, we did see replication lag spike to about 5-10 seconds, which can be an issue if you need to promote the secondary for read traffic. It's also crucial to regularly test your failover process; it can take a few minutes and there are DNS caching issues that the documentation doesn't cover extensively.

Just a heads up: with cross-region replication, there’s always some potential for async lag since data has to travel across the network. If high consistency is essential for your use case, you might want to look into options like DSQL (which is more synchronous but potentially slower). The best way to really know what works for you is to run tests tailored to your specific requirements.

One issue I've faced is with minor version updates. You need to remove the secondary cluster, update the primary, and then re-add the secondary, which can be time-consuming. Alternatively, there's a blue-green failover option, but it's still a bit clunky.

In my experience, the impact on performance is surprisingly minimal. The dedicated storage layer replication does a fantastic job of isolating the reader load. I primarily use PostgreSQL, but I've found Aurora to be generally excellent. One thing to watch out for, though, is that if your readers hit a certain workload threshold, they might be dropped to prevent WAL lag from increasing. I’m not sure if the same applies to MySQL. Overall, you shouldn't experience writer issues, but your reader performance can be somewhat limited in this setup.

You're spot on about minimal impact on the writer due to the storage-layer replication. We noticed only a slight increase in write latency—maybe 1-2%—which is pretty negligible. However, be mindful of the replication lag, especially during load spikes. Also, the failover process isn't instantaneous; it typically takes 1-2 minutes for planned failovers, which could be longer if there’s an unplanned event. If sub-30-second failover is a must, you might want to explore options like written forwarding.