Automation Toolsmithing

and Maintenance in Python

(For Acceptance Tests and Beyond)

Praveen Shirali

Test Architect, RiptideIO

BangPypers MeetUp - August 20th 2016, Bangalore, India.

This presentation IS NOT about:

  • How to write tests in python
  • Which test methodology to use
  • Which framework to use

This presentation IS about:

  • Building general-purpose tools for automation
  • Extending them to suite your product under test
  • Maintaining it as your product grows
  • Adding more such tools to your 'suite'

Layers of testing:

  • Unit tests
  • Acceptance tests
  • End to end tests
  • Performance tests
  • Scalability tests
  • ... and beyond!

In this talk we'll look at

Acceptance tests

and beyond...

Ground rules

  • Separate 'automation' from 'testing'
    automation = how you exercise functionality in an automated way
    testing = how you verify results
  • Automation code should go into a test-library
  • Testcase should comprise of test logic. It calls automation code, unaware of how it is performed.
  • Testcode and automation code command the same respect as product code.

Where to begin automation?

  • Identify the fastest way for your product to get to an isolated test-ready state. (process?, VM?, container?)
  • Identify interactions. (humans?, machines?)
  • Identify interface. (cli?, webAPI?, UI?, other I/O devices?)
  • Brainstorm use-cases, tests for best Return-On-Investment (ROI).

Your first wrapper library

  • Choose a library or a tool as a starting point.
    Example: For product test deployment and management. (subprocess?, vm-sdks?, docker?...)
  • Write a class with basic methods to wrap this library or tool for generic use.
    class Product(object):
    	def __init__(self, **config):
    		# consumes config based on which the product is launched
    	def start(self):
    		# starts the product; may be subprocess.Popen, may be docker run...
    	def is_alive(self):
    		# returns a bool on whether the product is running
    	def stop(self):
    		# stops the product
  • You've just built your wrapper lib. Version it. Test it. Maintain it.

Extending the wrapper library

  • Extend the wrapper to suit 'your' product by adding additional methods.
    Example: Your product stores logs, depends on a DB?
    class MyProduct(Product, MyDatabase):	# mixin
    	def __init__(self, **config):
    	def start(self):
    	def is_alive(self):
    	def stop(self):
    	def delete_log_files(self):	# implmemented by MyProduct
    	def reset_database(self):	# inherited from MyDatabase
    	def cleanup(self): 		# calls the above two methods

Mashing functionality based on common usage patterns

  • Build wrappers around functionality that can work together.
    Example: Mashup of a HTTP client and a response validator
    Given a route to an endpoint, the HTTP client knows which validator to invoke on the result. When a difference is found, it generates a diff of expected and actual responses.
  • API is simplified.
  • Only one place needs a change if a different functionality is desired.

APIs should always be simple -- hidden costs of APIs

  • Modification required at all places where an API is called. Prone to mistakes.
  • Hard to mentally track changes to APIs. Unlearning and relearning APIs is a pain.
  • API Documentation. Simpler APIs speak for themselves. Very little documentation required.

Choose your test framework wisely

  • Will your test code be python code or plain-text? (gherkin, DSL?)
  • You may be limited by features and bugs in what the test framework provides.
  • The test framework also decides the interface between the test library and the tests

Your test-library is king

  • Automation is not purely for tests alone. Think of simulations, long-running soak tests etc.
  • Consider an interactive python console to test-library APIs for quick hacks.
  • Handle CLI argument parsing through your test-library. Not through the test framework.
  • Launch the test-framework through your own script.

Avoid fixture hell

  • Fixture hell = mentally losing track of what a given fixture does
  • Keep fixtures close to your test code.
  • Pack common setup and teardown patterns into the test library.

Build a namespace of objects

  • You only need to deal with one object, which is a nested attribute tree to other objects.
  • You could use dependency-injection frameworks like SpringPython or just subclass from dict!
  • Test cases get immunity from changes to the test-library

Namespace example using dict -- Implementation

import json

class Dict(dict):

    def __init__(self, *args, **kwargs):
        super(Dict, self).__init__(*args, **kwargs)

    def __setattr__(self, attr, value):
        self[attr] = value

    def __getattr__(self, attr):
        return self[attr]

    def __hasattr__(self, attr):
        return True if attr in self else False

    def __repr__(self):
        return json.dumps(self, indent=4, sort_keys=True, default=repr)

Namespace example using dict -- Usage

d = Dict(one="ONE", two="TWO")
d.three = "THREE"
d.four = Dict(nested_four="NESTED_FOUR")

>>> print d
    "four": {
        "nested_four": "NESTED_FOUR"
    "one": "ONE",
    "three": "THREE",
    "two": "TWO"

>>> print d.four.nested_four

Namespace example using dict -- Extended

n = Namespace()
n.set("parent.child.grandchild", MyClass())

>>> n.parent
	"child": {
		"grandchild": {{instance of MyClass}}

>>> print d.four.nested_four

Rely on logging

  • Python logger is insanely awesome. Use it as much as possible.
  • Override the __repr__ in the classes you define. Let the instance tell you its data.
  • Testing is meant to be debug-friendly. Make every attempt to keep things readable.
  • Quick turn around time matters a lot with in-field issues. Prepare for it in advance.

Automated data generation

  • Use a library to generate planned or random test data.
    Example: hypothesis
  • When you know your data models, its easy to build equivalence classes and generate test data.
  • Parametrize data. Implement the code to generate data in your test-library.
  • Test with insanely large values, wrong datatypes, magic strings etc

And, always plan for the following ...

  • Parallelization of test execution (with replicated test environments)
  • Replacement of 1st line libraries which you wrapped initially.
  • Maintenance of the test library for multiple version of your product

... cool stuff you could consider doing

  • Use tagging. Build reports on features covered, issues fixed, verified etc.
  • Automatically scale testruns horizontally with additional resources.
  • Tag defect IDs. Pull defect status and automatically regress fixed issues.

Some essential reading

  • How to design a good API and why it matters Google Tech Talks, January 24, 2007
  • Zen and the art of automated test suite maintenance John Ferguson Smart, June 13, 2013
  • Automated Testing Patterns and Smells Google Tech Talks, March 6, 2008

Q&A, feedback...

Praveen Shirali