The value I found is being able to do multi-region read replicas with no compute overhead for lower traffic geos.

I like the idea of PS and have toyed around with the idea of migrating to it but there are some glaring issues I don’t want to deal with:

- no native way to export backups and avoid vendor lock in (or pay for the row reads to generate regular backups)

- contradictory cost model. Their pricing page reads “Every time a query retrieves a row from the database, it is counted as a row read. Every time a row is written to the database, it is counted as a row written.” while their docs state “Rows read is a measure of the work that the database engine does, not a measure of the number of rows returned. Every row read from a table during execution adds to the rows read count, regardless of how many rows are returned.”

It’s very common for scale-out architectures to read more data than is ultimately returned, because the former is pulled from individual shards and then some centralized filtering / post processing is applied in some API middleware layer.

Trying to fix that by pushing down more of the query/execution is sometimes but not always feasible or practical.

GP storage is pretty slow (relatively). I hope PS isn't giving us that out of the box and does provision more IOPs etc.

AWS bills for absolutely everything as PS mentioned in the article. This can include temporary files and what not that can take up a large chunk of your storage.

> because it doesn't specify the exact instance type used on the AWS side

That in a way is problematic because CPU generations matter and even if PS is using the "best" currently, how do we know if they'd keep up? Prices could swing the other way quite easily.