I too was impressed with SQL Server last time I used it (big note: SQL Server is one of the few commercial DB's that does real nested transactions; PG does not), but I get it.
More or less. The default ROLLBACK behaviour is to roll back the whole nest. You have to use SAVE TRANSACTION instead of BEGIN TRANSACTION and specify the name in ROLLBACK. If doing this in a procedure that may or may not be part of a nested transaction (an explicit transaction wasn't started before it was called) you have to test to see if you need to BEGIN or SAVE¹ and what to do if a rollback is needed (you likely don't want to ROLLBACK everything if you used SAVE, but have no option if you used BEGIN). Not exactly intuitive. And saved transactions can cause excessive lock escalation, impacting concurrent performance.
SQL Server is generally a damned fine product overall, both generally and compared to a lot of other things emitted by Microsoft, but it isn't even close to perfect in some areas.
----
[1] ref: https://learn.microsoft.com/en-us/sql/t-sql/language-element...
I'm not one to compliment Microsoft software products, but Analysis Services and SQL Server, at least when I used them at the time over 10 years ago, seemed like darn fine products (which they purchased from someone else, I believe...)
SQL Server v7 was a rewrite of significant parts of the internals, so from that point it is probably fair to call it an MS product quite distinct from Sybase's (which itself continued to be separately developed including significant but different improvements/reengineering of the internals), though still showing some signs of its heritage.
To the best of my knowledge.
Note: Am not a DBA, merely a full stack dev that has worked with various databases over the years.
https://gist.github.com/pmarreck/970e5d040f9f91fd9bce8a4bcee...
Pasted it there for brevity here. Also includes an easy way to export a Claude conversation (made with the help of Claude, of course).
Savepoints are fully equivalent to nested transactions with the constraint that only one is concurrently active. This limitation is just from the SQL standard programming model. At least in PostgreSQL implementation it would be reasonably simple to support multiple concurrent subtransactions. They are even called subtransactions internally.
It’s a key “primitive” missing from almost all high-level programming languages, with the notable exception of some SQL dialects.
Not sure where this myth keeps coming from, but no, it does not:
https://www.sqlskills.com/blogs/paul/a-sql-server-dba-myth-a...