HTTP streaming response support for Azure functions [In Public Preview Now]

[nitinl]

Is there support for real-time streaming HttpResponse from Azure function?

This question has been asked before, and the solution looks exactly like what is needed. However, func.HttpResponse object has no attribute 'write'.

[YunchuWang]

@nitinl thanks for following up! we are releasing a new worker version which supports http streaming req/response soon. stay tuned!

[rgn15996]

Any updates on ETA? Everyone wants to return nice streamed responses from LLMs!

[zhoxing-ms]

+1

[RoyceAroc]

+1

[DJSaunders1997]

+1

[eavanvalkenburg]

+1

[bastienjalbert]

Any news about current implementation status ?

[tyler-suard-parker]

@YunchuWang Sorry to bother you about this again, but my company's project is looking to stream response from an Azure functions app. Is that supported yet?

[fabge]

while you might have to wait a while until this is implemented, you could actually just use Azure Web PubSub as intermediary and stream responses this way.
e.g. a streaming response from the server could look like this:

1. user connects to websocket (Azure Web PubSub) using a /connect endpoint
2. user sends a message to /message endpoint
3. azure function receives the message and uses the azure pub sub sdk (e.g. for python) to stream the response via Azure Web PubSub to the user
4. user receives the message via the existing websocket connection

[waynebaby]

I am using .net, and i am confusing if i can use streaming output in .net version or it will ship with python features together?

[bastienjalbert]

It seems to be working with .net function already

[westleyMS]

Having issues with .net function streaming, seems I get response buffering causing issues. Can you point a link to a sample?

[RoyceAroc]

when is the expected release date?

[andredewes]

@YunchuWang Is there any rough estimate when this will be released?

[YunchuWang]

Hi everyone, thank you for your attention. We are actively working on supporting this feature for python function as one of top priorities.
The latest update on timeline is as below:

• Prototyping/Design completed internally
• PRs merged and ready for release.
• Public preview: around Microsoft build event time (May 21-23)

If there is any update on the timeline, i will share it here.

update1: feature is being released and will be in next azure function runtime release today
update2: we are preparing for public preview once released and will be out soon. will keep you updated.
update3: no need to sign up and instructions to enable http streaming will be given in public preview announcements along with samples
update4: it will be released into public preview around Microsoft build event time (May 21-23)
update5:
Hi everyone, thank you for the attention and interest. Great news, the http streaming feature is released in public preview today! Please feel free to check it out following the announcement https://azure.microsoft.com/en-us/updates/public-preview-azure-functions-support-for-http-streams-in-python/ and let us know your experience/feedback/issues :)!

EDIT: HTTP streaming is GA now! https://azure.microsoft.com/en-us/updates?id=492057

[linde12]

One eternity later... The (still, as of Feb 2025) beta of azurefunctions-extensions-http-fastapi still requires you to replace all func.HttpRequest to Request and it is also using fixed versions of pydantic so if your project uses pydantic you will have to settle with an outdated variant because of this extension package.

How is this even an acceptable solution for such a basic function?

Adding azurefunctions-extensions-http-fastapi pulls in not only fastapi but also anyio, pymongo, click, uvicorn and a bunch of other dependencies. Why do i need all these dependencies for such a basic feature?

Why not expose a write method or even better, allow the body to be a generator function (similar to the Node SDK where body can be a ReadStream)? Why does this have to be an extension, especially one relying on fastapi?

Great, i can store files in Azure Blob Storage but i cannot send them to the client because i would have to load gigabytes of data into memory before i construct & return a func.HttpResponse, instead of streaming chunks... 👍

[shahztechfarm]

I am facing the exact same issue. When I import
from azurefunctions.extensions.http.fastapi import Request, StreamingResponse
I get the error

[2025-02-17T18:13:55.941Z] Indexed function app and found 4 functions
[2025-02-17T18:13:55.945Z] Worker failed to index functions
[2025-02-17T18:13:55.946Z] Result: Failure
Exception: FunctionLoadError: cannot load the Function01 function: 'req' binding type "httpTrigger" and dataType "None" in function.json do not match the corresponding function parameter's Python type annotation "HttpRequest"

[yashchks87]

Same issue as mentioned above. Can you give update on this?

[s174182]

@yashchks87 I got the same experience with the same error message, it works for me if i switch my app from FastApi to eg:
`
import azure.functions as func
from FastApiApp import app as fastapi_app
app = func.AsgiFunctionApp(app=fastapi_app, http_auth_level=func.AuthLevel.ANONYMOUS)
´

To:
`
import azure.functions as func
from azurefunctions.extensions.http.fastapi import Request, StreamingResponse
import asyncio
import os
app = func.FunctionApp(http_auth_level=func.AuthLevel.ANONYMOUS)
´

[s174182]

Also, it seems that only "func.FunctionApp" supports actual streaming. func.AsgiFunctionApp still waits for the streaming to complete before returning a response.

Can someone elaborate whether above is correct and why such might be the case?

[vasudevan-n]

@YunchuWang Can i know if its supported for FunctionApps written in PowerShell ?