hi -

First off, it's hard for me to understand what you are claiming is a "leak" since your program runs just 100 iterations and exits. A "memory leak" is not simple growth of memory after running some operations, that's completely normal and SQLAlchemy has lots of internal structures that get built up when operations are first run, as well as an internal statement cache. So "run some operations for the first time, memory is now bigger" is perfectly normal.

this is how you show a "leak":

@profile
def main():

    async def amain(engine):
        async with engine.begin() as conn:
            await conn.run_sync(metadata.create_all)
        while True:
            async with AsyncSession(bind=engine) as session:
                s = await session.stream(select(None).where(TUser.id==1))
                [None async for __ in s]
                del s # just to show that the AsyncResult itself is not leaking
                await session.close()  #just to show the session is not leaking

note the "while True". Then you go into top and look at the memory use of the process overall. On this end, it's fixed and not moving at the values you see below (CPU 98% due to the tight loop):

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND                                                                                                          
2724788 classic   20   0  363464  61496  11980 R  98.3   0.8   1:02.61 python                                                                                                           

So I dont see this program indicating any kind of leak.

Beyond that, there is a real memory leak in greenlet with Python 3.11 that was fixed in Greenlet 2.0.1. Make sure you upgrade greenlet since you are using Python 3.11.

@zzzeek

"So "run some operations for the first time, memory is now bigger" is perfectly normal."

that's absolutely clear, but here I ran the exact same operation multiple times

"note the "while True". Then you go into top and look at the memory use of the process overall. On this end, it's fixed and not moving at the values you see below (CPU 98% due to the tight loop):"

on my machine the memory consumption was growing constantly until it crashed. Furthermore I would expect that at least some memory is released at latest when the session is closed, but that was not the case

Beyond that, there is a real memory leak in python-greenlet/greenlet#328 that was fixed in Greenlet 2.0.1. Make sure you upgrade greenlet since you are using Python 3.11.

Thanks. Here you saved me! After the update it works just fine

FYI I found I was using

greenlet==1.1.3.post0

I am sure it was installed as dependency with any other library I installed.

However, I think at least greenlet!=2.0 should also be included here:

all greenlet versions prior to 2.0.1 leak memory in Python 3.11, whereas on previous Python platforms, all greenlet versions work fine. the above line would have to be further qualified for python 3.11, but this gets into does a downstream package want to block libraries due to API incompatibilities (greenlet 0.4.17 in our case) or additionally for behavioral bugs (all greenlets prior to 2.0.1 on py311 only).

I think it would have to look like this

install_requires =
    greenlet >= 2.0.1;python_version>="3.11" and (platform_machine=='aarch64' or (platform_machine=='ppc64le' or (platform_machine=='x86_64' or (platform_machine=='amd64' or (platform_machine=='AMD64' or (platform_machine=='win32' or platform_machine=='WIN32'))))))
    greenlet != 0.4.17;python_version<"3.11" and (platform_machine=='aarch64' or (platform_machine=='ppc64le' or (platform_machine=='x86_64' or (platform_machine=='amd64' or (platform_machine=='AMD64' or (platform_machine=='win32' or platform_machine=='WIN32'))))))

which is not great