Agent gets stuck 'in-progress'

[anilaltuner]

Please read this first

• Have you read the docs?Agents SDK docs
  Yes
• Have you searched for related issues? Others may have faced similar issues.
  Yes

Describe the bug

I have an orchestrator agent and two hand‑off agents.
Everything behaves correctly when I run the stack on my local machine, but after deploying the exact same containers to AWS EC2 (Docker), one of the hand‑off agents stops responding. From the traces it looks like the request to POST /v1/responses hangs indefinitely.

Debug information

Item	Value
Agents SDK version	v0.0.12
Python version	3.11.x
Platform (local)	macOS / Docker Desktop
Platform (prod)	AWS EC2 t3.medium, Amazon Linux 2, Docker 24.0
OpenAI account / key	Same key in both environments

Expected Behaviour

It's weird because one of the agent works well but other is stuck with same environment.

On the other hand, from local, it works well.

[pakrym-oai]

Does the agent end up hanging indefinitely? The underlying SDK has a 10m timeout. Do you ever hit that?

Can you try a client with much shorter timeout:

set_default_openai_client(
    AsyncOpenAI(
        http_client=DefaultAsyncHttpxClient(timeout=httpx.Timeout(timeout=20, connect=5.0))
    )
)

?

[anilaltuner]

Yes it does, for example; my request from one day ago still says 'in progress' on tracing page.

Alright, I can try

[pakrym-oai]

my request from one day ago still says 'in progress' on tracing page.

That might be a result of the outer tracing span never being terminated because the process got killed or some other reason. The fact that you see POST /v1... means that the request finished because we only send the request span information to the server after it's done.

Is it possible you're getting an error in the application that causes it to exit?

[anilaltuner]

I did some deep dives and yes, it is about Django's thread safety features. Like you sad, when its stopped the working that happens.

Thanks for overall.