chain.run threw 403 error

[mzhadigerov]

I'm using the pipeline for Q&A pipeline on non-english language:

pinecone.init(
    api_key=PINECONE_API_KEY,  # find at app.pinecone.io
    environment=PINECONE_API_ENV  # next to api key in console
)
index_name = "langchain2"
embeddings = OpenAIEmbeddings(openai_api_key=OPENAI_API_KEY)

docsearch = Pinecone.from_existing_index(index_name=index_name, embedding=embeddings)

llm = OpenAI(temperature=0, openai_api_key=OPENAI_API_KEY)
chain = load_qa_chain(llm, chain_type="stuff")

query = "QUESTION"
docs = docsearch.similarity_search(query, include_metadata=True)

res = chain.run(input_documents=docs, question=query)
print(res)

It stucks in res = chain.run(input_documents=docs, question=query) lane. I'm waiting for ~20 mins already. What's the reason for that and how to investigate?

---

UPD

After ~30 mins I got

Retrying langchain.llms.openai.completion_with_retry.<locals>._completion_with_retry in 4.0 seconds as it raised Timeout: Request timed out: HTTPSConnectionPool(host='api.openai.com', port=443): Read timed out. (read timeout=600).

What's the reason for that and how can I fix it?

[dlqqq]

This looks like an issue with the upstream API and not LangChain itself. LangChain is not responsible for upstream API availability or rate limiting policies.

[mzhadigerov]

@dlqqq How can I set a limit for response time? I'm using signal library from python. Wondering If there are alternative ways

[dlqqq]

@mzhadigerov Unfortunately, a quick glance at the pinecone.py source seems to show that there isn't timeout support at the moment for Pinecone. The community is still working to bring timeouts to classes throughout the repo: #1798

[mzhadigerov]

sorry @dlqqq ,but what it has to do with pinecone? The problem arises in openai side, not in pinecone side. Please, correct me If I misunderstood you.

[dlqqq]

@mzhadigerov Sorry, I misread your error log. Yes, you are correct, this is an issue with the upstream OpenAI API.

How can I set a limit for response time?

There actually is a request_timeout option on the OpenAI providers. You can declare this when instantiating the provider:

TIMEOUT = 60 # <= timeout in seconds. default is 600 (set by OpenAI)
llm = OpenAI(temperature=0, openai_api_key=OPENAI_API_KEY, request_timeout=TIMEOUT)

This change was implemented by #398. Support for the request_timeout kwarg was implemented upstream in the OpenAI Python SDK here: openai/openai-python#116

[dlqqq]

@mzhadigerov Can we close this issue?