httpx.PoolTimeout, fixed by disabling HTTP keepalive

[charles-dyfis-net]

This is a deeply odd error, insofar as it's something that happens 100% of the time only when my code is invoking Anthropic (via client libraries 0.28.0) from our CI environment (hosted in AWS us-east-2).

We get a anthropic.APITimeoutError: Request timed out, thrown from a httpx.PoolTimeout, thrown from a httpcore.PoolTimeout during a generally unremarkable test case. The following change in usage/invocation prevents it entirely:

- async_client = AsyncAnthropic()
+ http_client = httpx.AsyncClient(
+     limits=httpx.Limits(max_connections=None, max_keepalive_connections=0)
+ )
+ async_client = AsyncAnthropic(http_client=http_client)

[RobertCraigie]

What version of httpx are you using?

If you're not on the latest I would strongly recommend upgrading, there were some issues with older versions of httpx not closing connections properly.

[charles-dyfis-net]

Indeed, this was observed when behind a few releases (httpx 0.25.2) -- I'll retest on 0.27.0 and see if that resolves the issue.

[charles-dyfis-net]

Unfortunately, the issue still reproduces with httpx 0.27.0.

[charles-dyfis-net]

Ah! Updating httpx alone didn't solve the problem.

I had a dependency pinning anyio, which prevented httpcore from being updated. Updating that pin to permit a modern anyio permits newer httpcore to be pulled in, and that resolves the problem.

Perhaps anthropic-sdk-python should explicitly specify appropriate minimum versions of both?

[rattrayalex]

Glad this problem was fixed for you! We don't want to pin when we can avoid it, as it can cause worse problems for people in some situations.