在 AWS Lambda 上使用 Tensorflow 部署微服务

Question

I have been stuck here for too long. I am trying to deploy a microservice that uses tensorflow. There is a single file by the name of handler.py which has the simple code below:

import json
import tensorflow as tf
import numpy as np

def main(event, context):
    # a = np.arange(15).reshape(3, 5)

    body = {
        "message": "Go Serverless v1.0! Your function executed successfully!",
        "input": event
    }

    response = {
        "statusCode": 200,
        "body": json.dumps(body)
    }

    return response

    # Use this code if you don't use the http event with the LAMBDA-PROXY
    # integration
    """
    return {
        "message": "Go Serverless v1.0! Your function executed successfully!",
        "event": event
    }
    """

To make my work easier I am using serverless to deploy a microservice but it fails saying, unzipped size is too big. Here is how my directory looks like:

-- handler.py
-- serverless.yml
-- requirements.txt

requirements.txt looks like:

numpy
tensorflow

I also tried to upload without installing the above modules thinking that lambda will itself initialize from requirements.txt but then get an error that Unable to import module 'handler': No module named 'tensorflow' . What should I do? I have spent lots of time in this and still not convinced that AWS Lambda would not allow me to do this.

If you wanted to see serverless.yml , it looks as follows:

service: numpy-new-test

provider:
  name: aws
  runtime: python3.6
  profile: nsp
  role: arn:aws:iam::xxxxxxxxxxx7:role/AdminRole

functions:
  numpy:
    handler: handler.main
    events:
      - http:
          path: test
          method: get 

Answer 1

As you've mentioned from the error you received it looks like your zipped package is too large. You received that other error because you have a module requirement in your script to use tensorflow.

Keep in mind the AWS Lambda Limits has a 50MB deployment package size limit. The Tensorflow package by itself is close to 50MB, so adding the Numpy package will take it well over the limit.

Have a look at this blog which does some investigation into the package limit sizing in AWS Lambda

https://hackernoon.com/exploring-the-aws-lambda-deployment-limits-9a8384b0bec3