What Happened Last Week

Battled through a problem with the imports on my Python 3 project. Spent time reading multiple definitive guides to solve the infamous Python “ModuleNotFoundError” exception. It didn't occur until I was 3 classes deep, so a unit test was referencing a class which imported another. My vlog this week is just on modules.

I got my log files saved to S3 buckets and I am quite happy because in the process I think I managed to reduce the complexity. The program was getting many IF statements because I was trying to switch between local hard disk storage during tests and S3 buckets. I started using Abstract Base Classes and everything looks and feels much better.

Python Package Structural Changes

The last 6 weeks had been going too smoothly, but within a week of upgrading to Python 3 I hit a very challenging problem. I wrote unit tests which worked but then when I ran the modules from the command line I would get import errors. Fix one way of running the program and the other wouldn't work.

After several days of intense reading the idea to look at some base code, I choose the calendar library and quickly realized you don't need to have one class per file the way I was used to doing it in C#.

Once I merge all my classes into one module/file, the ImportError: No module named XXXX basically, puff, disappeared. 

Actually, the solution looks much different to 6 weeks ago.

Feature Switch

class Log(object):
    """
    Stores latest execution information
    """

    FEATURE_ENABLED = False  # can be removed once S3 bucket is integrated

    def __init__(self, storage: AbstractLogStorage):
        self.storage = storage

    def latest_execution(self, job_name, price):
        if not self.FEATURE_ENABLED:
            return
        self._append_to_job_log(job_name, price)  # each job gets its own file with price history
        self._update_central_job_log(job_name, price)  # goes into the LAST_EXECUTED_FILENAME
...

Abstract Base Classes

In C# we get these plus Interfaces. I was searching for this early in the project when I started thinking about how to mock my S3 bucket and AWS Lambda functions.

class AbstractJobsStorage(ABC):
    """
    Jobs can be loaded from S3 or local hard disk (YML files)
    """

    @abstractmethod
    def load(self, object_name: str) -> dict:
        pass

Passing in (ABC) makes it abstract. You'll need to it to the imports list at the top of the file too.

from abc import ABC, abstractmethod

Then we can write different implementations. I started with the storing data locally, until I had the data structures sorted out.

class JobStorageOS(AbstractJobsStorage):
    """
    Load jobs from local hard disk
    """

    def __init__(self, filepath):
        self.filepath = filepath

    def load(self, object_name):
        with open(self.filepath + object_name, 'r') as logfile:
            return yaml.load(logfile)

The inheritance is set on the class using (AbstractJobsStorage).

In my unit test, I used it like this.

storage = LogStorageOS(self.STORAGE_PATH)
log = Log(storage)

If you haven't done it before, my explanation might not be enough. I suggest quickly clone the solution, run pip3 on the requirements.txt and requirements_unit_test.txt file, then nose2. If it works and you have the diagram you'll figure it. Also read the python.org doco on ABC too.

AWS Problems

Deploying Lambda functions doesn't mean they work. I'm thinking of writing a post-deployment test for that and mention it because it occurred a couple of times and I didn't spot it.

My solution has 2 AWS Cloud Formation templates, and I had to make modifications to the foundation template which sets up the environment. The Lambda functions needed additional permissions to write to S3 buckets. Pushing the changes to GitLab did not make them appear on AWS, and I realise now that I need to rework this part of the solution because I create part of the stack manually.

Figuring out the policy change for the role that my Lamba functions use was also tricky. I thought just adding the s3:PutObject and s3:ListBucket action directly to the template would be enough, but after it didn't work tried via the IAM policy editor to find out that s3:ListBucket needed additional restrictions. The good part here is you can copy paste back the policy into your Cloud Formation template.

On a final note, Lambda functions recently got a new feature called Layers. It allows functions to share libraries. I haven't bother implementing it yet.

Coming Next

I have setup AWS to trigger my web scraper every 2 hours and will let it run for a few days. In the meantime, I have already created a new project for the next phase of this project. To do what I did on Amazon but on Google Cloud Platforms (GCP).

https://gitlab.com/neilspink/gcp-price-grabber

I will be figuring out how to get the solution up and running on GCP.

What Happened Last Week

I've been working on my Python unit tests involving AWS Lambda functions. Those unit tests then became particularly handy, as I changed from using Python 2.7 to Python 3.6. Finally, I also started work on a class for creating log files but that is unfinished.

The source code can be found here https://gitlab.com/neilspink/aws-lambda-price-grabber

Unit Test Code Coverage

Knowing what % of the source code gets tested can help you identify parts of a system you forgot to unit test and happened to me while developing AWS Lambda functions that were calling other Lambda functions. I started searching how to get the code coverage on my project and found there is a --with-coverage command line switch for nose2 utility I've been using.

nose2 --with-coverage

I increased the code coverage from 26% to 65% the invoker class which was calling other AWS Lambda functions. The increase is good enough for me. I'm not a believer in having high percentage coverages, because all too often the tests are less functional. I prefer if I can learn something about a program from a unit test, i.e. a unit test setting and reading string parameters is not valuable, doing an action or calculation is. 

Unit Test Mocking

I'm not sure if you know what mocking is, but in case you haven't heard this term, it is just a way to replace parts of your system under test with mock ones, faking part of the system to simulate the behaviour of the real ones. For me, I needed to mock AWS Lambda functions. 

I lost a lot of time trying a library called Moto which I found on GitHub, it looked very promising, but I gave up on it. I spent several hours trying to get it to work, my final test before I gave up was cloning the library and running nose2, and none of the unit tests worked on my computer. 

Luckily I found the documentation on Python.org 😀 https://docs.python.org/3/library/unittest.mock.html, and I got my unit tests working with the patch. I said this at the beginning 5-weeks ago; I don't think you need to pay for any online courses, the documentation is all there. Although I don't find it easy to read sometimes, it was worth the effort.

Mocking an AWS Lambda Function

A little background on what I was testing. In my web scraper project, I want to get the prices from multiple websites. I have a list of jobs and invoke the grabber to get the prices.

My first step, I moved all the code from my Lambda function to a class named Invoker.

from invoker import Invoker

def lambda_handler(event, context):
    if 'source' not in event:
        raise Exception("The 'source' key is missing from the event dictionary.")

    job = Invoker()

    result = job.grab(event)

    print(result)
    return result

   def grab(self, event):
        ...
        for site in website_list['sites']:  
            ...
            response = self._invoke_lambda(payload)

@staticmethod
def _invoke_lambda(payload):

    client = boto3.client('lambda')

    return client.invoke(FunctionName='grab-price',
                         InvocationType='RequestResponse',
                         Payload=payload)

In just wanted it to return an arbitrary text message "none". For that I added at the top

from mock import patch

and used the patch function decorator

lambda_result = "none" 

with patch.object(Invoker, '_invoke_lambda', return_value=lambda_result):
    result = Invoker().grab(events)

Coming Next

I want to finish saving prices that get scrapped from websites, schedule my crawler and watch it run for a couple of days.