Upgrade aiohttp

[Dreamsorcerer]

No description provided.

[Dreamsorcerer]

Database credentials don't work anymore? I haven't touched those in the PR..

[Dreamsorcerer]

@NateBrady23 or anyone know what to change for the DB credentials? Would be good to get this upgrade in as there's been a lot of performance optimisations in recent releases (though they may be a magnitude more noticeable in real world scenarios than these simple benchmarks).

[volyrique]

Could it be an issue with the authentication method (issue #8061 might be related)?

[Dreamsorcerer]

Could it be an issue with the authentication method (issue #8061 might be related)?

I'm unclear. We just create a DSN with user/password, so I don't see anything we'd need to change for that:

FrameworkBenchmarks/frameworks/Python/aiohttp/app/main.py

Line 32 in 1beae77

password=os.getenv('PGPASS', 'benchmarkdbpass'),

Appears that asyncpg implemented the auth method 6 years ago, so I doubt it's an issue there either..

[volyrique]

Yes, but the point is that the password is not necessarily sent in cleartext to the server - see here. asyncpg might not be using the correct method (the defaults have changed much more recently than 6 years ago, as the issue I linked to demonstrates, and some methods might have been removed - other frameworks have been broken by database updates), hence failing with an authentication error. The error message doesn't seem to go into that kind of details, so I would probably run a packet capture to see what is happening on the wire, unless asyncpg provides that information somehow (I am not familiar with it).

Just an idea, since you were asking for suggestions.

[msmith-techempower]

The issues here don't SEEM to be related to the md5 issue... closing for now, but feel free to open a new PR when you get the issues resolved.

[Dreamsorcerer]

I don't have the time to figure out and test locally what is going wrong with your test suite, nothing has changed on the code side..
Given that the problem is already present, and yet it appears results have been included in the latest benchmark run, I would assume that you have some way to run this successfully, regardless of the CI result.

[msmith-techempower]

It doesn't run successfully locally or in CI with your changes https://github.com/TechEmpower/FrameworkBenchmarks/actions/runs/13909801752/job/38921326418?pr=9348

Currently, on main aiohttp verifies successfully.

This may be related to #7557 since the only changes that would result in the password authentication failing given your changes appear to be the db driver versions being updated. Maybe these new versions disabled md5 auth?

[Dreamsorcerer]

Ahhh, I didn't realise it was still working on master. Apparently sqlalchemy changed rendering of URL objects.

This should be good to merge now.