Asynchronous processing

Deploying AI into Production with FastAPI

Matt Eckerle

Software and Data Engineering Leader

What is asynchronous processing

A person serving a drink to someone sat at a table

Allows handling multiple requests concurrently

Synchronous vs asynchronous requests

A chart showing difference in processing times for synchronous and asynchronous requests

Turning synchronous endpoints asynchronous

@app.post("/analyze")
def analyze_sync(comment: Comment):
    result = sentiment_model(comment.text)
    return {"sentiment": result}

import asyncio

@app.post("/analyze")
async def analyze_async(comment: Comment):
    result = await asyncio.to_thread(
        sentiment_model, comment.text
    )
    return {"sentiment": result}

Implementing background tasks

from fastapi import BackgroundTasks
from typing import List

@app.post("/analyze_batch")
async def analyze_batch(
    comments: Comments,
    background_tasks: BackgroundTasks
):

    async def process_comments(texts: List[str]):
        for text in texts:
            result = await asyncio.to_thread(
              sentiment_model, text)
    background_tasks.add_task(process_comments, 
                              comments.texts)
    return {"message": "Processing started"}

BackgroundTasks manage comment processing queue

background_tasks handles post-response processing.

add_task schedules process_comments asynchronously.

Adding error handling

@app.post("/analyze_comment")
async def analyze_comment(comment: Comment):
    try:
        sentiment_model = SentimentAnalyzer()
        result = await asyncio.wait_for(
            sentiment_model(comment.text),
            timeout=5.0
        )
        return {"sentiment": result["label"]}

Adding error handling

    except asyncio.TimeoutError:
        raise HTTPException(
            status_code=408,
            detail="Analysis timed out"
        )

    except Exception:
        raise HTTPException(
            status_code=500,
            detail="Analysis failed"
        )

Let's practice!

Deploying AI into Production with FastAPI