Trace Init duration with FastAPI

[CGarces]

Hi!

This is more a lambda + fastAPI question than aws-lambda-web-adapter itself but...

After read #283 my expectation was fast cold start below 2 seconds. I have a lambda with a cold start between 2500 ms and 2800 ms

A typical cold start, when point to /api/account in order to check id the user is authenticated.

REPORT RequestId: 0b863c3f-9f47-4486-b248-2b0366d08cda Duration: 2.47 ms Billed Duration: 3 ms Memory Size: 512 MB Max Memory Used: 141 MB Init Duration: 2735.78 ms

Just 3ms to handle the user data but 2735.78 ms to start the lambda.

The same request on warm lambda.

REPORT RequestId: d93c662f-047a-4c8a-863c-1d043d58837f Duration: 2.43 ms Billed Duration: 3 ms Memory Size: 512 MB Max Memory Used: 142 MB

How can I trace what is goin on during the 2.7 seconds of the cold start?
Any hint to optimize fastAPI startup without provisioned concurrency?

Simplified python code using .zip package.

from fastapi import Depends, FastAPI, status
from fastapi.responses import JSONResponse
from common import is_user_login, oauth_manager
import boto3
import fastapi
import api
import os


app = FastAPI(docs_url="/api/docs/", redoc_url=None)
oauth_manager.attach_to_app(app)
# Objeto que controla la conexión del usuario
app.include_router(api.router)


client = boto3.client("s3")


@app.get("/api/account")
async def account(
    user_info=Depends(
        oauth_manager.get_user_info_from(["header", "cookie"], auto_error=False)
    ),
    isLogged=Depends(is_user_login),
):
    if user_info:
        return JSONResponse(...)


@app.get("/management/info")
async def info():
    return JSONResponse(...)


@app.post("/uploadurl")
async def get_upload_url(name: str = fastapi.Body(...)):
     url = client.generate_presigned_post(...)

    return url

[bnusunny]

You could add -X importtime option to python startup command. This will log how much time is spent on importing each package.

[CGarces]

import time: self [us] | cumulative | imported package
import time: 1145 | 1002282 | fastapi
import time: 206330 | 1080114 | common

I don't know the scale of the numbers this microseconds?

1 second for fastapi, 1 second for my onw logic...
I have fixed the problem implementing 304 responses and "stale-while-revalidate" cache control for cloudfront.

[bnusunny]

Yes, the unit is microseconds.