Making simultaneous calls with images to FASTapi using request-async library in python not working

Question

Following is the sample code making 10 calls to the API.

URL = "http://127.0.0.1:3000/v1/img_process"

CALLS = 10

payload = {"data": "value1, value2, value3, value4, value5, value6"}
files = {'file1': open('sample1.jpg', 'rb'),
        'file2': open('sample1.jpg', 'rb')}


headers = {
    "Authorization" : "Bearer XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"
}

async def request_post():
    return await requests.post(url=URL,headers=headers, data=payload, files=files) 

async def main():
    start = datetime.now()
    tasks = [asyncio.create_task(request_post()) for _ in range(CALLS)]
    result = await asyncio.gather(*tasks)

    end = datetime.now()

    print("Start time = " + start.strftime("%H:%M:%S") + "\n" + 
        "End time = " + end.strftime("%H:%M:%S") + "\n" +
        "Elapsed = " + str(end - start) + "\n" + 
        "Seconds per call = " + str((end - start)/CALLS)
        )


await main()

Out of the 10 calls only the 10th call(last) is being executed by the API, in the first 9 cases the API is throwing the following error

PIL.UnidentifiedImageError: cannot identify image file <_io.BytesIO object at 0x000002115FA50E00>

The code for image reading at the endpoint is

def read_file_as_image(byte_data) -> np.ndarray:
    image = np.array(Image.open(BytesIO(byte_data)))
    return image 

@app_v1.post("/img_process")
async def inputs(data: List[str] , file1: UploadFile = File(...), file2: UploadFile = File(...)):

img1 = read_file_as_image(await file1.read())
img2 = read_file_as_image(await file2.read())

Could this be the issue with using the the async/await function while reading the images Please suggest some solution

Answer 1

Have you looked into FatsAPI BackgroundTask?

https://fastapi.tiangolo.com/tutorial/background-tasks/

You can add the task read_file_as_image() in BackgroundTask as it takes some time to process image.

It also could be that the function read_file_as_image is not truly async and multiple calls cancel execution and last call is only executed fully.