DocumentIntelligenceClient.begin_analyze_document and poller.result() block event loop despite async API

[danielbkang]

• Package Name: azure-ai-documentintelligence
• Package Version: 1.0.0 - 1.0.2
• Operating System: Windows 11, Linux (Azure Functions Python 4.x Docker base image)
• Python Version: Python 3.11

Describe the bug
The begin_analyze_document method of DocumentIntelligenceClient (from the async azure.ai.documentintelligence.aio package) and its returned poller’s result() method perform blocking I/O internally. Despite being part of the async SDK, these calls block the asyncio event loop, causing delays and concurrency issues in frameworks like FastAPI.

To Reproduce
Steps to reproduce the behavior:

1. Use DocumentIntelligenceClient from azure.ai.documentintelligence.aio in an async FastAPI endpoint.
2. Call await client.begin_analyze_document(...).
3. Await poller.result().
4. Observe that the FastAPI event loop is blocked during the call, causing other requests to stall.

Expected behavior
Both begin_analyze_document and poller.result() should perform non-blocking async I/O, allowing other asyncio tasks to run concurrently without delay.

Screenshots
N/A

Additional context
As a temporary workaround, offloading the blocking calls to a thread using asyncio.to_thread() avoids event loop blocking.

[github-actions]

@bojunehsu @vkurpad

[bojunehsu]

@catalinaperalta Can you please help take a look and see if there are underlying issues in Azure core?

[catalinaperalta]

Tagging @xiangyan99 @kashifkhan to weigh in on the lro behavior here.

[kashifkhan]

Hi @danielbkang, are you able to provide some sample code that can repro this scenario for us?

Also a few follow up questions:

• Are you using any custom transports and using that instead of the default aiohttp based one.
• Are you using any specific tools to determine that the eventloop is blocked ?

[github-actions]

Hi @danielbkang. Thank you for opening this issue and giving us the opportunity to assist. To help our team better understand your issue and the details of your scenario please provide a response to the question asked above or the information requested above. This will help us more accurately address your issue.

[danielbkang]

Repro Code

import base64
from azure.ai.documentintelligence.aio import DocumentIntelligenceClient
from azure.ai.documentintelligence.models import AnalyzeDocumentRequest
from azure.core.credentials import AzureKeyCredential

async def extract_text_as_markdown(document_analysis_client: DocumentIntelligenceClient, file_bytes):
    poller = await document_analysis_client.begin_analyze_document(
        model_id="prebuilt-layout",
        body=AnalyzeDocumentRequest(bytes_source=base64.b64encode(file_bytes).decode("utf-8")),
        output_content_format="markdown"
    )
    result = await poller.result()
    return result

# Called from Azure Function Queue Trigger
async def main(msg: QueueMessage) -> None:
    async with DocumentIntelligenceClient(endpoint=endpoint, credential=AzureKeyCredential(key)) as client:
        extracted_text = await extract_text_as_markdown(document_analysis_client=client, file_bytes=file_bytes)

• Are you using any custom transports and using that instead of the default aiohttp based one.

No, I am not using any custom transports — this uses the default aiohttp-based transport.

• Are you using any specific tools to determine that the eventloop is blocked ?

No, I didn’t use specific tools to detect event loop blocking.

However, I diagnosed the issue in the following way:

In Azure Functions: when this function runs, unrelated endpoints(FastAPI +Azure Function) freeze while awaiting the result about 6–7 seconds, just before poller.result() completes.

As a workaround, I replaced the async client (azure.ai.documentintelligence.aio.DocumentIntelligenceClient) with the sync version and ran the same logic inside asyncio.to_thread(). When I did this, the stalling behavior disappeared.