Are you happy with Aurora Serverless so far? Is it really the "painless scaling" it's made out to be? I'm really hesitant to use anything other than self-hosted Postgres with full control, because I'm always afraid of running into situations like these and not being able do what I need to do, like have an offline analytical read replica. Also I imagine the cost is an order of magnitude more than a self-hosted instance.
Not really. But it's mostly because the support for other features is basically non-existant. It also seems like all development is happening on Aurora Serverless v2 (which I think is in preview state?). The scaling also isn't seamless, it breaks your connections if you're using prepared statements - we have it pinned to a specific number of capacity units.
We'll probably move to plain Aurora at some point, as that supports all bells and whistles, but for now Aurora Serverless is good enough.
