I have some questions, I don't have any suggestions yet, but after some discussion we might have the right answer and add to this document, or upcoming updates.

1. Any idea how we could resolve situation that some databases have multiple different drivers. In the same Hook or in the different one? e.g postgres have 3 quite popular (sync implementation)

• psycopg2 - More common driver nowadays, use libpq
• psycopg - formally psycopg3 but all future versions v4, v5, and etc. would be released as part of this package
• pg8000 - python implementation of postgres protocol, do not required/use libpq

2. Some drivers also implements both interfaces DBAPIv2 compliant and "internal" which usual faster, and have bigger functionality, what our preferences here?

3. Not all DB have sqlaclhemy support. However in the current implementation of DbApiHook.get_sqlalchemy_engine. Any thought how we could make it more clear? And maybe we should remove default implementation and raise NotImplemented instead?

4. Broken get_uri mechanism in DBAPIHook. And right now it also not clarify what kind of URI it expected to get: driver or sqlalchemy? What we should do to improve it? For the reference and details: https://lists.apache.org/thread/8rhmz3qh30hvkondct4sfmgk4vd07mn5

5. Async implementation of drivers, this is extending of question 1. Most of the drivers supports either sync or async but not both, e.g. psycopg have support both but in addition also quite popular asyncpg or aiopg. Should we have separate hook for async drivers or we try to extend DbApiHook and make it really big an complicated?

I have some questions, I don't have any suggestions yet, but after some discussion we might have the right answer and add to this document, or upcoming updates.

1. Any idea how we could resolve situation that some databases have multiple different drivers. In the same Hook or in the different one? e.g postgres have 3 quite popular (sync implementation)

• psycopg2 - More common driver nowadays, use libpq
• psycopg - formally psycopg3 but all future versions v4, v5, and etc. would be released as part of this package
• pg8000 - python implementation of postgres protocol, do not required/use libpq

2. Some drivers also implements both interfaces DBAPIv2 compliant and "internal" which usual faster, and have bigger functionality, what our preferences here?
3. Not all DB have sqlaclhemy support. However in the current implementation of DbApiHook.get_sqlalchemy_engine. Any thought how we could make it more clear? And maybe we should remove default implementation and raise NotImplemented instead?
4. Broken get_uri mechanism in DBAPIHook. And right now it also not clarify what kind of URI it expected to get: driver or sqlalchemy? What we should do to improve it? For the reference and details: https://lists.apache.org/thread/8rhmz3qh30hvkondct4sfmgk4vd07mn5
5. Async implementation of drivers, this is extending of question 1. Most of the drivers supports either sync or async but not both, e.g. psycopg have support both but in addition also quite popular asyncpg or aiopg. Should we have separate hook for async drivers or we try to extend DbApiHook and make it really big an complicated?

Yep. I think those are great questions and I propose to add and discuss separate ADR entry for each of those questions - this is precisely what ADRs are for - separate ADR for each such decision. I think what we did with comon.sql was just "basic" commonalisation" and we shoudl continue with all the points you raised. I will merge it now. And we can discuss each of thsoe separately and propose ADRs for them individually.