AWS Lambda and Supplychainpy (there was an attempt)

I recently attempted to use supplychainpy from an AWS lambda function, and it did not go well. The problem was not the library itself or the process required to create a python deployment package, with all the environment dependencies, and uploading to AWS. I document some of the process below. This post is not a tutorial but I suppose it might be useful as confirmation of how not to go serverless.

Building the Deployment Package

I used Anaconda on an Ubuntu Linux machine to build the virtual environment for deploying the package. AWS offers Python 3.6 and 2.7 for building lambda functions. Using the conda command, I built an environment for python 3.6:

$ conda create -n suchpy_lambda python=3.6

After creating and activating the new environment, I installed the dependencies (Cython for my project and Pillow and boto3 for AWS):

$ pip install Pillow boto3 Cython

I then installed ‘supplychainpy’ (0.0.5rc0):

$ pip install supplychainpy==0.0.5rc0

To package all the dependencies I navigated to the anaconda site-packages for the environment created earlier:

$ cd ~/anaconda/envs/suchpy_lambda/lib/python3.6/site-packages

I zipped the dependencies found in the virtual environments site-packages and moved the compressed file to a project directory:

$ zip -r9 ~/Projects/python/supplychainpylambda/scmlambda.zip *

After creating scmlambda.zip and moving it to a project directory. I created a scmlambda.py file and added the following function:

from __future__ import print_function
import boto3
import json as js
import os
import sys
from decimal import Decimal
from supplychainpy import model_inventory

def handler(event, context):
    data_set = event['data_set']
    data_set = dict(js.loads(data_set))
    z_value = Decimal(event['z_value'])
    lead_time = Decimal(event['lead_time'])
    holding_cost_percentge = Decimal(event['holding_cost_percentage'])
    retail_price = Decimal(event['retail_price'])
    unit_cost = Decimal(event['unit_cost'])
    reorder_cost = Decimal(event['reorder_cost'])
    quantity_on_hand = Decimal(event['quantity_on_hand'])
    backlog = Decimal(event['backlog'])
    sku_id = event['sku_id']
    currency = event['currency']
    uncertain_demand = model_inventory.analyse_orders(data_set=data_set,
                                                      sku_id= sku_id,
                                                      lead_time= lead_time,
                                                      reorder_cost= reorder_cost,
                                                      z_value = z_value,
                                                      retail_price= retail_price,
                                                      unit_cost= unit_cost,
                                                      currency=currency,
                                                      quantity_on_hand=quantity_on_hand
                                                     )
   return uncertain_demand
  
if __name__ == "__name__":
	handler()
	    

This function from the supplychainpy library provides a summary of basic descriptive statistics and inventory KPI’s. The aim was to use an API Gateway as a trigger and send values to the function using JSON. As I explain below, I didn’t even get the chance to see if this specific function would work.

Now the newly created python file is added to the previously created zip file:

zip -g scmlambda.zip scmlambda.py

At this point, all seemed to be going well. I had read elsewhere that you might want to do all of this on an AWS server instance. So I may have faced problems even if I had succeeded in uploading the deployment package.

To upload the Python deployment package, I selected an AWS blueprint, specifically the hello-world-python3.

Once you have chosen a blueprint, named it and selected a role, you choose a code entry type. I uploaded from an S3 instance, which requires copying the link to the location of the file over to the Lambda function.

It was at this stage I realised my package was too big to upload directly as the maximum is 10mb, so you are required to upload to an S3 bucket.

I received the error message that read:

The Code tab failed to save. Reason: Unzipped size must be smaller than 262144000 bytes:

I believe my failure underscores the point of a lambda function and how my attempt flies in the face of that logic. Bundling all the functionality of into a library that the function can use is not a sensible choice. It would mean each lambda would have a version of the library etc.

I am still experimenting with AWS lambdas and the library but my initial attempt to use it lump sum in a single function was not a success. It would make more sense to separate all the library functionality across different lambda functions. I might give that a try when I have the chance.