I know that you currently have yarl pinned to yarl==1.9.4 and aiohttp pinned to aiohttp==3.10.2 for testing, but in Fedora we use the system versions of all dependencies.
To reproduce this in a virtualenv, I ran the tests manually, which is a bit messy, but the below steps work. I had to unpin zipp to fix a dependency conflict. I started with a “control” run:
@musicinmybrain thanks for reporting, fyi if you install tox you can run tox -e py312-test-instrumentation-aiohttp-client instead of installing each dependency manually
Describe your environment
OS: Fedora 42/Rawhide (development version) Python version: 3.13.0 (or 3.12.6) Package version: 1.27.0 (or current
main
, e4ece57a8136310ed379c04fa77c6b6f8d009bbc)What happened?
I know that you currently have
yarl
pinned toyarl==1.9.4
andaiohttp
pinned toaiohttp==3.10.2
for testing, but in Fedora we use the system versions of all dependencies.This is a follow-up to https://github.com/open-telemetry/opentelemetry-python-contrib/issues/2916, which I closed after I discovered that the original regression from
yarl
1.13.0 was fixed by updatingaiohttp
to 3.10.7 or later.Steps to Reproduce
To reproduce this in a virtualenv, I ran the tests manually, which is a bit messy, but the below steps work. I had to unpin
zipp
to fix a dependency conflict. I started with a “control” run:(All tests passed for both packages.)
Now, updating
yarl
andaiohttp
:Expected Result
Actual Result
Additional context
I was able to pin a similar issue in
aioflo
, https://github.com/bachya/aioflo/issues/67, down toaiohttp
3.10.6, so I suspect that may be where the relevant behavior change occurred for this package as well. See https://github.com/aio-libs/aiohttp/blob/v3.10.6/CHANGES.rst#3106-2024-09-24.The tests for
opentelemetry-instrumentation-aiohttp-server
succeed with the latestyarl
andaiohttp
.Would you like to implement a fix?
None