-
Notifications
You must be signed in to change notification settings - Fork 4.4k
Closed
Labels
bugSomething isn't workingSomething isn't working
Description
Confirm this is an issue with the Python library and not an underlying OpenAI API
- This is an issue with the Python library
Describe the bug
A weird issue that I am facing in production, after the code is deployed and it has been 24-26 hours the any calls made using the above client start timing out. Usually taking out a redeployment helps.
This pattern makes it feel like the underlying HTTP connections are not being released or refreshed correctly, which may be causing a connection pool exhaustion.
Any best practices or recommendations on how to tackle this ?
My configuration
openai.api_key = os.environ.get("OPENAI_API_KEY")
os.environ.setdefault("OPENAI_LOG", "debug")
client = AsyncOpenAI(
http_client=DefaultAioHttpClient(
limits=httpx.Limits(max_connections=1000, max_keepalive_connections=100)
),
)
I am streaming and once i am done with streaming I am closing the stream as well
response = return await chat_client.chat.completions.create(**params)
## Later on
if response:
await response.close()
name = "openai"
version = "1.99.9"
To Reproduce
- NA since it it not easily replicable in local.
Code snippets
OS
Debian Linux
Python version
3.10.6
Library version
1.99.9
Metadata
Metadata
Assignees
Labels
bugSomething isn't workingSomething isn't working