Sentry Answers>FastAPI>

FastAPI StreamingResponse not streaming with generator function

FastAPI StreamingResponse not streaming with generator function

David Y.

—

The Problem

For one of the endpoints in my FastAPI, I need to return a large amount of data. I want to use FastAPI’s StreamingResponse so that my API’s client applications can start processing the data as it arrives. I’ve built a simplified prototype to test out the idea:

Click to Copy
from fastapi import FastAPI
from fastapi.responses import StreamingResponse

app = FastAPI()

def data_generator(): # simulate data streaming
    for number in range(100):
        yield f"{number}\n"

@app.get("/stream")
def stream_data():
    return StreamingResponse(data_generator())

But, when I query the /stream endpoint in my browser, I receive the entirety of the data in one response instead of as a stream. Why isn’t this working as I expected and how can I fix it?

The Solution

This issue is likely caused by the way browsers buffer responses – the client may be buffering the entire stream before displaying it as a single response. To force browsers and other HTTP clients to treat our response as a stream, we need to set its Content-Type to text/event-stream.

It’s possible that the data_generator function finishes too quickly for the data to be streamed in more than a single chunk. We should add a time delay to the function to better simulate a real data stream.

Here is an updated version of the code with these two changes applied:

Click to Copy
from fastapi import FastAPI
from fastapi.responses import StreamingResponse
import time

app = FastAPI()

def data_generator():
    for number in range(100): # simulate data streaming
        yield f"data: {number}\n\n"
        time.sleep(1)  # 1 second delay to simulate streaming

@app.get("/stream")
def stream_data():
    return StreamingResponse(data_generator(),
                              headers={ "Content-Type": "text/event-stream" }) # appropriate Content-Type header added to response

When moving beyond this simple prototype and manual testing in the browser, it is important to ensure that the HTTP client consuming this request supports HTTP streaming.

  • Syntax.fmListen to the Syntax Podcast
  • Community SeriesIdentify, Trace, and Fix Endpoint Regression Issues
  • ResourcesBackend Error Monitoring 101
  • Syntax.fm logo
    Listen to the Syntax Podcast

    Tasty treats for web developers brought to you by Sentry. Get tips and tricks from Wes Bos and Scott Tolinski.

    SEE EPISODES

Considered “not bad” by 4 million developers and more than 100,000 organizations worldwide, Sentry provides code-level observability to many of the world’s best-known companies like Disney, Peloton, Cloudflare, Eventbrite, Slack, Supercell, and Rockstar Games. Each month we process billions of exceptions from the most popular products on the internet.

learn more about sentryjoin our discord
© 2024 • Sentry is a registered Trademark of Functional Software, Inc.