4 ms·

I spotted this over the weekend, nice work zzzeek! The technical details still evade me somewhat but it seems we can now run a big blob of synchronous code with

by aidos 6y ago

I spotted this over the weekend, nice work zzzeek! The technical details still evade me somewhat but it seems we can now run a big blob of synchronous code with async around the edges. How deep are the changes required to sqlalchemy? Do you just need to add some wrappers around going out to asyncpg? EDIT I'm interested in this because the last time I looked, running sqla with socketio looked like it wasn't going to be fun. Then my words were "if I can't have sqla, then what's the point?" - so we used Hasura to do the bulk of the donkey work instead. Boy do I miss sqla when I'm in that part of the system though.

zzzeek 6y ago
the changes are so minimal it's almost embarrassing. basically none within SQLAlchemy or anything else. you run in a greenlet and everything is normal. when you get down to the driver, it's asyncpg, oh OK then we use await_(). if you look at https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/2071 https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/2071, and scroll down, the Gerrit display shows you a listing of files. next to each file is a bar that is a combination of green and red. Green means lines that are added, red means lines that are removed. at the bottom, you can see the total score, as of revision 13 it's 1581 lines added, 90 lines removed. That is an insane ratio meaning literally nothing was changed, the vast majority of the code is the new asyncpg dialect where I had to replicate the DBAPI and a lot of other details that DBAPIs usually provide up front. it's been several days now and it's looking more and more like the issue of switching between asyncio and blocking style code is really easy to solve, unless I'm in some kind of hallucinatory fugue state for the past three days and all of these passing tests aren't really there.

4 ms·

I spotted this over the weekend, nice work zzzeek! The technical details still evade me somewhat but it seems we can now run a big blob of synchronous code with

by aidos 6y ago

zzzeek 6y ago
the changes are so minimal it's almost embarrassing. basically none within SQLAlchemy or anything else. you run in a greenlet and everything is normal. when you get down to the driver, it's asyncpg, oh OK then we use await_(). if you look at https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/2071 https://gerrit.sqlalchemy.org/c/sqlalchemy/sqlalchemy/+/2071, and scroll down, the Gerrit display shows you a listing of files. next to each file is a bar that is a combination of green and red. Green means lines that are added, red means lines that are removed. at the bottom, you can see the total score, as of revision 13 it's 1581 lines added, 90 lines removed. That is an insane ratio meaning literally nothing was changed, the vast majority of the code is the new asyncpg dialect where I had to replicate the DBAPI and a lot of other details that DBAPIs usually provide up front. it's been several days now and it's looking more and more like the issue of switching between asyncio and blocking style code is really easy to solve, unless I'm in some kind of hallucinatory fugue state for the past three days and all of these passing tests aren't really there.