In this lesson, Scott presents a diagram and I was under the impression that when we replicate on Multi-AZ we have to use Stand-by, and when we do it on Cross-Region, we have to use Read Replicas. Thus, can’t we have a read replica in the same region? Is a Stand-by instance by default a Read Replica? Or the Stand-by instance is NOT used unless as in a failover strategy? I guess I understand the different goals for a Stand-By (sync) and a Read Replica (async), but not so sure about the possibility of having both in the same instance… Or what happens if the "leader"/main node fails and I only have a read replica IN THE SAME region… Actually, can I have a read replica even in the same AZ (just for allowing all application’s query screens point to it, instead of to the "leader"/main node)?
2 Answers
So, here is what I’ve found: 1) https://aws.amazon.com/rds/features/multi-az/?nc1=h_ls and 2) https://aws.amazon.com/rds/aurora/faqs/… In summary, both maintain a copy of the database but use Multi-AZ deployments for High Availability/Failover (synch replication) and Read Replicas for reading scalability/basic availability (asynchronous replication) in and across regions;
Additionally, and apparently, stand-by instances are NOT used until a problem/fail-over situation (differently than a read replica). And, to make it clear, you could have read replicas in the same or in different regions, but they will be always under asynchronous replication (but with very low latency – milliseconds).
Aside from what you found, going to a low level in storage, if you are doing a total in-sync operation, you will need to wait for the acknowledge of it between the nodes.
That means, that if you configure a sync replication between two different regions, you will need to add the total time the operation is being written in X systems and you will need to receive the confirmation from all of them before confirming that the operation is done.
That is why you should keep any sync target [Stand-By, in this case] in close vicinity of the master server and why the graph shown by Scott makes sense.
I’m glad that you were able to find the answers to your own question and I hope I was able to add something to it.
Cheers!