I’ve gotten a few questions about Aurora Serverless v2 preview, so here is what you need to know about Aurora Serverless v2 during preview.

Q : Does it replace the existing Aurora Serverless offering?

A : No, it lives side-by-side with the existing Aurora Serverless, which will still be available to you as “v1”.

Q : Aurora Serverless v1 takes a few seconds to scale up, that’s too much for our use case where we get a lot of spikes. Is that the same with v2?

A : No, v2 scales up in milliseconds, during preview the max ACU is only 32 though.

Q : Is the cold start time for Aurora Serverless v2 still a few seconds?

A : Yes, unfortunately…

Q : So if you want to avoid cold starts, what’s the minimum ACU you have to run?

A : Minimum ACU with v2 is 0.5.

Q : Does v2 still scale up in double increments, e.g. 4 ACU -> 8 ACU?

A : No, it scales up in increments of 0.5 ACUs, so it’s a much tighter fit for your workload, so you’ll waste less money on over-provisioned ACUs.

Q : Is there anything I can do with v2 that I can’t do with v1?

A : Yes, v2 supports all the Aurora features, including those that v1 is missing, such as global database, IAM auth and Lambda triggers.

Q : Wait, but it’s twice as much per ACU!

A : Yes, but v1 requires a lot of over-provisioning because it doubles ACU each time and takes 15 mins to scale down. v2 scales in 0.5 ACU increments and scales down in < 1 min. AND you get all the Aurora features!

Q : Can you use “provisioned” and “serverless” instances in the same Aurora cluster?

A : Yes, you can! It’s pretty cool, right!?

Q : Is the HTTP Data API supported on v2?

A : Not in the preview, I guess it’ll be there when Aurora Serverless v2 becomes GA.

Q : If using from Lambda, do I need to use RDS Proxy to manage the connections to the cluster? Data API kinda mitigated that for v1.

A : Yes, you probably should, at least until Data API is enabled on v2. Like with Aurora Serverless v1, more connections requires more ACUs. So if there’s a spike in traffic and Lambda scales up the number of concurrent executions then you can run into trouble with other maxing out the available connections, or it’ll hurt you economically because you will need much more ACUs because of the number of connections.

Jeremy Daly has written a nice summary post on Aurora Serverless v2 as well, with a nice experiment on its scaling behaviour. Go check it out if you haven’t already.