Posts Tagged ‘Continuous Integration’

Automating Specification by Example for .NET Web Applications

February 22, 2014

If you or your organisation:

  1. are/is constrained to running your .NET tests (unit, acceptance) on-site rather than in the cloud
  2. would like some guidance on how to set-up Continuous Integration

read on.

Introduction

Purpose

Remember, an acceptance test system as a tool is only as good as the specification provided by it’s humans. The most important ingredients there-for is the relationships between the people creating the tests and the interactions performed by those people. Or as the Agile Manifesto states: Value “Individuals and interactions over processes and tools”. In order for an acceptance test system to be successful, the relationships of the Developers creating the increment and the interactions between them and the stake holders must be in good shape first. Once this is in order, you can take the next step and find some tools that will assist in creating working software that does what the stake holders want it to do.

It’s my intention that the following details will help you to create a system that automates “Specification by Example”.

The purpose of providing an automated Specification by Example Implementation, A.K.A Automated Acceptance Test System, is clearly explained here.

Do not fall into the trap of inverting the test triangle. Instead invest where it matters.

Scope

Create a system that can be triggered from

  1. Every developers workstation
  2. A build on the build machine, preferably from a best of bread build tool. TFS is not a best of bread build tool and if you want to get serious about Continuous Integration (CI), nightly builds, continuous deployment, I’d recommend not going down the path of TFS. Even Microsoft uses Git. Doesn’t that tell you something? Do you see TFS here? Last time I evaluated build tools, Jenkins previously named Hudson came out on top.

jenkins

The system will include

  1. An acceptance test framework that will run all the acceptance tests
  2. A Unit test framework. UI tests need to be run in parallel on a collection of VM’s (See the section on supported browsers for why). There are three immediately obvious approaches we could take here.
    1. We could try and rely on a unit test framework to distribute the tests. MSTest 2012 doesn’t provide the ability to run tests in parallel, but 2010 does. In order to have 2012 run tests in parallel, you can force it to use the 2012 test settings file. Only a maximum of 5 tests can be run concurrently though. Not a great option, considering it’s not going to be supported going forward.
    2.  My ParallelBrowser. If this link is not active and you’re interested in this, contact me.
    3. PNUnit. An example of how this works is here under the “PNunit Framework for writing selenium test cases” heading. I wrote the ParallelBrowser before Selenium had good support for running the same tests on multiple supported browsers. Both my ParallelBrowser and this option are reasonable options, but I’d go for the latter now. This way someone else can maintain the parallel aspect. As unless people are interested in ParallelBrowser I won’t be doing any further work on it.
  3. A Web User Interface Test Framework that will be driven by the acceptance test framework. Selenium in this case.
  4. A set of tests that run Selenium tests. These will of course need to be thread-safe.
  5. As per the Supported Browsers section, a collection of VM’s with our supported browsers installed.
    1. Each with a standalone selenium server setup with a role of webdriver. Details further on.
  6. A stand-alone selenium server setup with a role of hub

High Level Flow

Many organisations bound to .NET seem to be locked into using sub-standard tooling like TFS for their build. If you are in this predicament and can not break free, I’d suggest once all the unit tests, integration tests have run, then have the build kick off a psake script to:

  1. Clean out the existing target web app
  2. Deploy the newly built and tested web app
  3. Drop the database
  4. Create database by using latest DDL and DML scripts pulled from source control
  5. Apply any specific configurations
  6. Stop and start the target web server
  7. Run the acceptance tests which will include any Web UI tests.

If it’s within your power to choose a real CI Tool to run in-house, there are a handful of very solid contenders. A good proportion of which are free and open source.

Audience

Who ever is setting up the system. Often a developer or two. It’s important to make sure more than one person knows how it all hangs together, otherwise you have a single point of failure.

Chosen Tools

Evaluation Criterion I used

  • Who is the creator? I favour teams rather than individuals, as individuals move on often leaving projects stranded?
  • Does it do what you need it to do?
  • Does it suite the way you and your team want to work?
  • Does it integrate well with all of your other chosen components? This is based on communicating with those that have used the offerings more so than using Proof Of Concepts (POC).
  • Works with the versions of dependencies you currently use.
  • Cost in money. Is it free? Are there catches once you get further down the road? Usually open source projects are marketed as is. No catches
  • Cost in time. Is the set-up painful? Customisation feedback? Upgrade feedback?
  • How well does it appear to be supported? What do the users say?
  • Documentation. Is there any / much? What is its quality?
  • Community. Does it have an active one? Are the users getting their questions answered satisfactorily? Why are the unhappy users unhappy (do they have valid reasons).
  • Release schedule. How often are releases being made? When was the last release?
  • Intuition. How does it feel. If you have experience in making these sorts of choices, lean on it. Believe it or not, this should probably be No. 1

The following tools have been my choice based on the above criterion.

Acceptance Test Framework

The following offerings are all free and open source.

If you’re not using User Stories and/or Test Conditions, the context/specification offerings provide greater flexibility than the xBehave style frameworks. As most Scrum teams use User Stories for their Product Backlog items and drive their acceptance tests with test conditions, xBehave offerings are a great choice. In saying that, there is probably no reason why both couldn’t be used where it makes sense to do so. In this section I’ve provided the results of evaluating the current xSpec and xBehave offerings for .NET ordered by best first for the categories.

xBehave (test conditions)

SpecFlow

specflow

  • Sourcecode: https://github.com/techtalk/SpecFlow/
  • Age: Over 4 years
  • Actively maintained: Yes
  • Large number of active committers
  • Community: Lively
  • Visual Studio Plug-in has been downloaded 70 times as many times as NBehave
  • Documentation: Excellent
  • Integrates well with Selenium (I’ve setup a couple of systems using SpecFlow and it’s been a joy to work with). The stake holders loved the visibility it provided too. I discussed it here in a recent presentation.
NBehave
  • Not a lot of activity
  • Only two committers
StoryQ
  • Only two coordinators
  • Well established framework

xSpec (context/specification)

Machine.Specification (MSpec)
NSpec

Web User Interface Test Framework

selenium

For me when I look at this category of tools for .NET, Selenium is always at the top and it just keeps getting better. If anyone has any questions around Selenium, feel free to contact me or leave a comment on this post. I can’t guarantee I’ll have the answer, but I’ll try. All the documentation can be found here. I would recommend installing the Selenium IDE for initially recording tests and be sure to check-out the IDE plug-ins. All the documentation you’ll need for the IDE is here. Once you get familiar with the code it generates, you will not use it much. I would recommend using the newer Web drivers rather than the selenium server by itself. The user group is very active and looks like a good place to ask questions also. Although I haven’t needed to as there is a huge amount of documentation that’s great.

The tools I would use are detailed here. Specifically we would be using

  1. Selenium 2 (aka WebDriver)
  2. The IDE for recording tests initially
  3. Selenium Server which is used by WebDriver and RC (now considered legacy) now includes built-in grid capabilities.

Supported Browsers

What I’ve done in the past is have each of our supported versions from each supported browser vendor installed on a single VM. So each VM has all the vendors browsers installed, but just a single version obviously.

Mid Level Flow

These are the same points listed above under “High Level Flow

1. Build Kicks off PSake Script

psake

The choice to use PSake over the likes of NAant, Rake and the other build scripting languages is reasonably straight forward for me. PSake (PowerShell build scripting language) gives us access to the full .NET environment. NAnt with all it’s angle brackets, was never a very nice scripting language to use. Rake is excellent and a possible option if you have ruby installed. If you don’t, why install it if you have .NET? There are many resources for PowerShell on the inter-webs. The wiki for PSake is good.

In the case where you may have a TFS Build run, I would suggest once all the unit tests and integration tests have run, then the build kicks off a possibly pre-build and post-build psake script to perform the following operations. This is how you do this. Oh, before you try to actually run a PSake script, download and import the module, or install the NuGet package. So once you have your PSake scripts running, just start adding PowerShell scripts to do the following work. PSake is just syntactic sugar around PowerShell, so anything you can do with PS, you can do with PSake.

2. Clean out the existing target web application

Using your PSaki script, use the Web Deploy cmdlets. You will find everything you need here for it. You can also install the NuGet package.

3. Deploy the newly built and unit tested web application

As above, just use the Web Deploy cmdlets.

4. Drop the database

As above, just use the Web Deploy cmdlets.

5. Create database by using latest DDL and DML scripts pulled from source control

Database update via Application

Kind of related, but not specific to CI.

Depending on your needs, there are quite a few ways you could do this.

One way of doing this is to have your application utilise a library that determines which version of the database the application needs and be able to update the database accordingly. This library would use similar or the same upgrade scripts that we would use in this test process.

Your applications should create (if non existent) and update database on run. So all the DDL, DML code per database lives in a library. Each application that uses a specific database, references the databases DDL code library. Script all stored procedures, views, functions, triggers they’re recreated as part of a deployment scrip.

When the application is deployed, and the database created or updated, anything that must be there for the application to run out of the box should be part of the scripts, and of course versioned. This includes the part of our data that is constant or configuration data. Tables, stored procedures, views, functions and triggers. For the variable part of your data, you will need a synthetic data generation plan for testing.

Database Process for Versioning

Also related, but not specific to CI.

DBA, Devs, Product Owner and consultants must be aware of the process.

When any schema, constant data, configuration data, test data is updated… the (version controlled) scripts must also be updated, else the updates will get overwritten.

As part of the nightly build, if your supporting multiple versions of your application, you could also hydrate the collection of database versions, then run the appropriate upgrade scripts against each one, to verify the upgrades work. If any don’t, the build fails.

Create set of well defined processes that:

  1. In most cases, looks after itself
  2. Upgrades existing databases if they are not on the latest version, to the latest version
  3. Creates databases for those applications that don’t have a database
  4. Informs the user on deployment if the database is corrupt, or can not be upgraded
  5. Outlines who is responsible for, and who may update the DDL and DML scripts for your projects
  6. Clearly documents that any changes made to any databases by un-authorised personal will more than likely be overwritten.

A User Story for this might look something like the following:

As the team, we need to create a set of well defined processes that clearly outline what is required in regards to setting up the development teams database versioning, creation, upgrade systems and processes strategy for our organisations databases. So that all team personal are aware of the benefits and dangers of making changes to the databases, and understand the change process.

Possibly useful tools

1. DB Ghost
2. http://www.red-gate.com/products/sql-development/sql-source-control/index-2
3. http://www.sqlaccessories.com/SQL_Data_Examiner/

6. Apply any specific configurations

As above, just use the Web Deploy cmdlets.

7. Stop and start the target web server

As above, just use the Web Deploy cmdlets.

8. Run the acceptance tests which will include any Web UI tests

As above, just use the Web Deploy cmdlets.

  1. Start each VM that hosts a set of browsers you want to use to farm your tests out to. From memory, you do not need to start each browser. There are of course many ways to do this. PS provides the following cmdlets Start-VM and Stop-VM. These would be my first options.
  2. Start the selenium standalone server. All details found here. Or just work through the “Distributed Testing with Selenium Grid” chapter until you get to the “Creating and executing Selenium script in parallel with TestNG” heading, at which point switch to this documentation to replace TestNG with PNUnit.

If I’ve failed to explain anything in enough detail for you, drop me a message below and I’ll do my best to help 🙂

Advertisements

How to optimise your testing effort

March 24, 2012

I recently wrote a post for the company I currently work for around the joys of doing TDD.
You can check it out here.

What is your current approach to testing?
How can you spend the little time you have on the most important areas?

I thought I’d share some thoughts around where I see the optimal areas to invest your test effort.
I got to thinking last night, and when I was asleep.
We are putting too much effort into our UI, UA and system tests.
We are writing to many of them, thus we’re creating a top heavy test structure that will sooner or later topple.
These tests have their sweet spot, but they are slow, fragile and time consuming to write.

We should have a small handful for each user story to provide some UA, and the rest should be without the UI and database (the slow and fragile bits).
We need to get our mind sets lower down the test triangle.

test triangle

I’ll try and explain why we should be doing less Manual tests, followed by GUI tests, followed by UA tests, followed by integration tests, followed by Unit tests.

Try not to test the UI with the lower architectural layers included in the tests.
UI tests should have the lower layers mocked and / or stubbed.
Check out Dummy vs Fake vs Stub vs Mock
Full end to end system tests are not required to validate UI field constraints.
Dependency injection really helps us here.

When you are explicitly testing the upper levels of the test triangle, the lower / immediate lower layers are implicitly being tested.
So you might think, cool, if we invest in the upper layers, we implicitly cover the lower layers.
That’s right, but the disadvantages of the higher level tests outweigh the advantages.
UI tests and especially ones that go from end to end, should be avoided, or very few in number,
as they are fragile and incur high maintenance costs.
If we create to many of these, confidence in their value diminishes.
Read on and you’ll find out why.

Lets look at cost vs value to the business.

Some tests cost a lot to create and modify.
Some cost little to create and modify.
Some yield high value.
Some yield low value.
We only have so much time for testing,
so lets use it in the areas that provide the greatest value to the business.
Greatest value of course, will be measured differently for each feature.
There is no stock standard answer here, only guidelines.
What we’re aiming for is to spend the minimum effort (cost) and get the maximum benefit (value).
Not the other way around…
With the following set of scales, we’ve spent to much in the wrong areas, yielding suboptimal value.

cost verse business value

It’s worth the effort to get under the UI layer and do the required setup incl mocking the layers below.
It’s also not to hard to get around the likes of the HttpContext hierarchy of classes (HttpRequest, HttpResponse, and so on) encountered in ASP.NET Web Forms and MVC.

Beware

  • the higher level tests get progressively more expensive to create and maintain.
  • They are slower to run, which means they don’t run as part of CI, but maybe the nightly build.
    Which means there is more latency in the development cycle.
    Developers are less likely to run them manually.
  • When  they break, it takes longer to locate the fault, as you have all the layers below to go through.

Unreliable tests are a major cause for teams ignoring or losing confidence in automated tests.
UI, Acceptance, followed by integration tests are usually the culprits for causing this.
Once confidence is lost, the value initially invested in the automated tests is significantly reduced.
Fixing failing tests and resolving issues associated with brittle tests should be a priority to remove false positives.

Planning the test effort

This is usually the first step we do when starting work on a user story,
or any new feature.
We usually create a set of Test Conditions (Given/When/Then)

Given When Then
There are no items in the shopping cart Customer clicks “Purchase” button for a book which is in stock 1 x book is added to shopping cart. Book is held – preventing selling it twice.
Customer clicks “Purchase” button for a book which is not in stock Dialog with “Out of stock” message is displayed and offering customer option of putting book on back order.

for Product Backlog items where there are enough use cases for it to be worth doing.
Where we don’t create Test Conditions, we have a Test Condition workshop.
In the workshop we look at the What, How, Who and Why in that order.
The test quadrant (pictured below) assists us in this.
In the workshop, we write the previously recorded Acceptance Criteria on a board (the What) and discuss the most effective way to verify that the conditions are meet (the How)
With the how we look at the test triangle and the test quadrant and decide where our time is most effectively spent.

Test condition workshop

With the test condition workshop,
when we start on a user story (generally a feature in the sprint backlog),
we plan where we are going to spend our test resource.
Think about What, and sometimes Who, but not How.
The How comes last.

Unit tests are the developers bread and butter.
They are cheap to create and modify,
and consistently yield not only good value to the developers,
but implicitly good value to most / all other areas.
This is why they sit at the bottom of the test triangle.
This is why TDD is as strong as it is today.
test quadrant

The hierarchy of criteria that we use to help us

  1. Release Criteria
    Ultimately controlled by the Product Owner or release manager.
  2. Acceptance Criteria
    Also owned by the Product Owner.
    Attached to each user story, or more correctly… product backlog item.
    The Development team must meet these in order to fulfill the Definition of Done.
  3. Test Conditions
    When executable, confirm the development team have satisfied the requirements of the product backlog item.

Write your tests first

TDD is  not about testing, it’s about creating better designs.
This forces us to design better software. “Testable”, “Modular”, separating concerns, Single responsibility principle.
This forces us down the path of SOLID Principles.

red green refactor

  1. Write a unit test
    Run it and watch it fail (because the production code is not yet written)
  2. Write just enough production code to make the test pass
  3. Re-run the test and watch it pass

This podcast around TDD has lots of good info.

Continuous Integration

Realise the importance of setting up CI and nightly builds.
The benefit of having your unit (fast running) tests automatically executed regularly are great.
You get rapid feedback, which is crucial to an agile team completing features on time.
Tests that are not being run regularly have the risk that they may be failing.
The sooner you find a failing test, the easier it is to fix the code.
The longer it’s left unattended, the more technical debt you accrue and the more effort is required to hunt down the fault.
Make the effort to get your tests running on each commit or push.

Nightly Builds

The slower running tests (that’s all the automated tests above unit tests on the triangle), need to be run as part of a nightly build.
We can’t have these running as part of the CI because they are just too slow.
If something gets in the way of a developers work flow, it won’t get done.

Pair Review

Don’t forget to pair review all code written.
In my current position we’ve been requesting reviews verbally and responding with emails, comments on paper.
This is not ideal and we’re currently evaluating review software, of which there are many offerings.