Asynchronous calls sometimes hanging - transitive httpcore issue since 1.x?

[cmcconomyfwig]

Hi,

We have been using httpx successfully for some time; in the last release, we upgraded from 0.25.0 to 0.25.1, and we started to experience some situations where we have one of our services calling another one of our services. These are python azure functions deployed to azure, if it matters.

What we saw was that the first service made a request, the second service completed its response (based on all logs), but the first service never seemed to react to that completion, and held open until our service's 30 minute kill timer hit and shut down the whole instance.

What I noticed was that httpcore moved up to 1.x from 0.18.x in your dependencies, so I dug around there.

I found that httpcore 1.x was published oct 6, and this issue showed up a few weeks later, which sounds related:
"AsyncConnectionPool hangs future connections sometimes when a connection get canceled"
encode/httpcore#830

I think you may have inherited a bug - I would recommend rolling back your dependency until this is figured out.
Personally I am pinning your app to 0.25.0 and will be keeping an eye on httpcore to see what happens next there.