The Challenges of BIG Testing: Automation, Virtualization, Outsourcing, and More

MB Full-day Tutorials

5/5/2014 8:30:00 AM

The Challenges of BIG

Testing: Automation,

Virtualization, Outsourcing,

and More

Presented by:

Hans Buwalda

LogiGear

Brought to you by:

340 Corporate Way, Suite 300, Orange Park, FL 32073

888-268-8770 ∙ 904-278-0524 ∙ [email protected] ∙ www.sqe.com

Hans Buwalda LogiGear Hans Buwalda has been working with information technology since his high school years. In his thirty year career, Hans has gained experience as a developer, manager, and principal consultant for companies and organizations worldwide. He was a pioneer of the keyword approach to testing and automation, now widely used throughout the industry. His approaches to testing, like Action Based Testing and Soap Opera Testing, have helped a variety of customers achieve scalable and maintainable solutions for large and complex testing challenges. Hans is a frequent speaker at STAR conferences and is lead author of Integrated Test Design and Automation: Using the Testframe Method.

4/10/2014

1

© 2014 LogiGear Corporation. All Rights Reserved

Hans Buwalda

LogiGear

Automation, Virtualization,

Outsourcing, and More

STAREAST 2014, Tutorial MBOrlando, Monday May 5, 2014

The Challenges of

BIG Testing


Introduction

− industries

− roles in testing

4/10/2014

2


About LogiGear

� Software testing company, around since 1994

� Testing and test automation expertise, services and tooling− consultancy, training− test development and automation services− "test integrated" development services

� Aims to be thought leader, in particular for large and complex test projects

� Products:− TestArchitect™, TestArchitect for Visual Studio™− integrating test development with test management and automation− based on modularized keyword-driven testing

www.logigear.comwww.testarchitect.com


About Hans

� Dutch guy, living and working in California since 2001, as CTO of LogiGear

� Background in math, computer science, management

� Original career in management consultancy, since 1994 focusing on testing and test automation− keywords, agile testing, big testing, . . .

www.happytester.comhans @ logigear.com

4/10/2014

3


Topics for today

� Automation

� Designing and organizing tests

� Executing tests

� Team, organization and process

� Off-shoring, globalization


What is "BIG"

� Big efforts in development, automation, execution and/or follow up

� It takes a long time and/or large capacity to run tests (lot of tests, lot of versions, lot of configurations, ...)

� Scalability, short term and long term

� Complexity, functional, technical, scale

� Number and diversity of players and stakeholders

� Various definitions of "big" possible... and relevant...− "10 machines" or "10 acres"− "1000 tests" or "1000 weeks of testing"

� Big today means: big for you− not trivial, you need to think about it

"Windows 8 has undergone more than 1,240,000,000 hours of testing"

Steven Sinofsky, Microsoft, 2012

4/10/2014

4


Existential Questions

� Why test?

� Why not test?

� Why automate tests?

� Why not automate tests?


Why test?

� People expect us to do

� Somebody wants us to

� Increases certainty and control − Showing absence of problems

� Finds faults, saving time, money, damage− Showing presence of problems

4/10/2014

5


Why not test?

� It costs time and money

� You might find problems . . .

� We forgot to plan for it

� We need the resources for development

� It is difficult

� It's hard to manage


Why Automate Tests?

� It is more fun

� Can save time and money− potentially improving time-to-market

� Can capture key application knowledge in a re-usable way

� Consolidates a structured way of working− when established as integral part of system

development process

� Can speeds up development life cycles

� Execution typically is more reliable− a robot is not subjective

4/10/2014

6


The Power of Robot Perception

FINISHED FILES ARE THE RE

SULT OF YEARS OF SCIENTI

FIC STUDY COMBINED WITH

THE EXPERIENCE OF YEARS...


Why not Automate?

� Can rule out the human elements− promotes "mechanical" testing− might not find "unexpected" problems

� More sensitive to good practices− pitfalls are plentiful

� Creates more software to manage

� Needs/uses technical expertise in the test team

� Tends to dominate the testing process− at the cost of good test development

maintenance can crush automation...

4/10/2014

7


Olny srmat poelpe can raed tihs.I cdnuolt blveiee taht I cluod aulaclty uesdnatnrd waht I was rdanieg. The phaonmneal pweor of the hmuan mnid, aoccdrnig to a rscheearch at Cmabrigde Uinervtisy, it deosn't mttaer in waht oredr the ltteers in a wrod are, the olny iprmoatnt tihng is taht the frist and lsat ltteer be in the rghit pclae. The rset can be a taotl mses and you can sitll raed it wouthit a porbelm. Tihs is bcuseae the huamn mnid deos not raed ervey lteter by istlef, but the wrod as a wlohe.

The Power of Human Perception


The Power of Human Perception

Notice at an event:"Those who have children and don't know it, there is a nursery downstairs."

In a New York restaurant:"Customers who consider our waiters uncivil ought to see the manager."

In a bulletin:"The eighth-graders will be presenting Shakespeare's Hamlet in the basement Friday at 7 PM. You are all invited to attend this drama."

In the offices of a loan company:"Ask about our plans for owning your home."

In the window of a store:"Why go elsewhere and be cheated when you can come here?"

4/10/2014

8


About tests in big projects

� Regular tests may be activities, complex tests are products. In fact any test that you want to run more than once is a product

� Every test that is written down with sufficient detail should be automated

� Automation− No longer an option in most situations− Also a key prerequisite of most agile approaches

� How tests are written and automated can make or break large scale testing


Keywords (actions) to help scalability

� The test developer creates tests using keywords

� I call them "actions", with each action having a keyword and arguments

� The automation task focuses on automating the keywords

� Each keyword is automated only once

3 actions, each with an action keyword and arguments

read from top to bottom

prod id name stock

new product hamm hammer 5

prod id quantity

add to stock hamm 20

prod id expected

check stock hamm 25

4/10/2014

9


Potential benefits of keywords

� More productive: more tests, better tests− more breadth− more depth

� Easier to read and understand

− no program code, tests can be self-documenting

− facilitates involvement of non-technical people, like domain experts

� Fast, results can be quickly available− the design directly drives the automation

� More targeted efforts for the automation engineers− less repetition, test design helps in creating a structured solution (factoring)− can focus on critical and complex automation challenges more easily

� Automation can be made more stable and maintainable− limited and manageable impact of changes in the system under test

� A significant portion of the tests can typically be created early in a system life cycle− dealing with execution details later

� . . .


Case: International Financial Project

� One of the largest projects to date with action words

� Over 10000 windows, meant for use in 85 countries

� Long development cycle (400 pp, 4 years and counting)

� Maintenance very hard

� Testing major bottleneck

� Much investment in automation techniques were needed to become successful

� Also a lot of attention for team and work environment helped the success

� Team of 35 test developers, 2 automation engineers

4/10/2014

10


Risks of keywords

� Keywords are often seen as silver bullet − often treated as a technical "trick", complications are underestimated

� The method needs understanding and experience to be successful− pitfalls are many, and can have a negative effect on the outcome− some of the worst automation projects I've seen were with keywords

� Testers might get pushed into half-baked automation role− risk: you loose a good tester and gain a poor programmer− focus may shift from good (lean and mean) testing to "getting

automation to work"− the actual automation challenges are better left to a the experienced

automation professionals

� Lack of method and structure can risk manageability− maintainability may not as good as hoped− tests may turn out shallow and redundant


Case: A Financial Project

� Large project, with a large automation team

� Early adopter of the keywords idea

� However, the keywords were applied with little method or direction

� Management did not see a need for help/coaching

� Result: many hard to maintain tests− crushing complexity− redundant tests and actions− the method was abandoned

4/10/2014

11


Keywords need a method

� By themselves keywords don't provide much scalability− they can even backfire and make automation more cumbersome− a method can help tell you which keywords to use when, and how to

organize the process

� Today we'll look at Action Based Testing (ABT)− addresses test management, test development and automation− large focus on test design as the main driver for automation success

� Central deliveries in ABT are the "Test Modules"− developed in spreadsheets− each test module contains "test objectives" and "test cases"− each test module is a separate (mini) project, each test module can

involve different stake holders


Example of an ABT test module

� Consists of an (1) initial part, (2) test cases and (3) a final part

� Focus is on readability, and a clear scope

� Navigation details are avoided, unless they're meant to be tested

TEST MODULE Car Rental Payments

user

start system jdoe

TEST CASE TC 01 Rent a car

first name last name car weeks

enter rental Mary Renter Ford Escape 2

last name amount

check payment Renter 140.42

FINAL

close application

4/10/2014

12


Example of a "low level" test module

� In "low level" tests interaction details are not hidden, since they are the target of the test

� The right level of abstraction depends on the scope of the test, and is an outcome of your test design process

TEST MODULE Screen Flow

user

start system john

TEST CASE TC 01 "New Order" button

first name control

click main new order

window

check window exists new order

FINAL

close application


Re-use actions to make new actions

� In the below example we use another sheet, but if you code actions, you could do something similar

� Often low level tests are re-used into these action definitions

ACTION DEFINITION check payment

user default value

argument last name Jones

argument amount

window control value

enter main last name # last name

window control

click main view balance

window control expected

check main balance # amount

4/10/2014

13


High Level Test Design - Test Development Plan

Objectives

Test Module 1

Test Cases

Test Module 2 Test Module N

Actions

. . .

AUTOMATION

Objectives Objectives

interaction test business test

Overview Action Based Testing

define the "chapters"

create the "chapters"

create the "words"

make the words work

Test Cases Test Cases


enter log in user name jdoe

enter log in password car guy

window control property expected

check property log in ok button enabled true

user password

log in jdoe car guy

first last brand model

enter rental Mary Renter Ford Escape

last total

check bill Renter 140.42


Case: Stock Exchange

� Transition from floor-based to screen-based trade

� Created on basis of an existing standard package− result: very little specifications

� Consisting of four major, different, systems that need to work in real-time

� Failures and bugs are not an option:− core of the financial system of the country, 100K revenue per second− traders not necessarily following rules

� In-depth knowledge limited to four people− nicknamed "The Four Daltons", after characters in a French comic book series about the

wild west− none of the four Daltons was involved in testing, testing was in a vacuum

� Three months to go...− test development (and scripted automation) had failed− test department not cooperating well with developers and domain experts− internal and external auditors had raised the alarm− and... the Dutch Crown Prince was scheduled to put the system into use!!

The Four Daltons(French comic book characters)

4/10/2014

14


Case: Stock Exchange

� Test set:− make it comprehensive− make it in-depth and aggressive− make it easy to assess and approve

� Organization:− get the right people involved (testing, automation, etc)− use scarce resources efficiently (in particular the four Daltons)− work with stake holders to let the process be transparent

� Technical:− use of the keyword method ("action words")− use "test objectives" so auditors can see quickly what you're testing− use great test design, don't mix apples and oranges

� "Sign off lubrication":− auditors signed off on the tests, not the test results− "the test is complete", not "the system works well"

� Results:− deadline was met one day before final date− the automated tests were the only ones used for acceptance− no functional errors found afterwards


Question

� What is wrong with the following pictures?

4/10/2014

15


No Y2K Problems in Auckland Airport??


Anything wrong with this instruction ?

You should change your battery or switch to outlet power immediately to keep from losing your work.

4/10/2014

16


Why Better Test Design?

� Many tests are often mechanical now− blindly follows specs or reqs− which often suits ok, but lacks aggression− no combinations, no unexpected situations− "methodical" does not have to mean "mechanical"

� For a higher “ambition level” you need− understanding of the system under test, and the business under test− analytical understanding of what could go wrong− creativity, and the commitment to use it

� Poor test development results in− cumbersome automation due to lack of focus− tedious retest cycles, loosing the agile advantage

Are you suffering from lame tests

too?


Effects of Better Test Design

� Quality and manageability of test − many tests are often quite "mechanical" now− one to one related to specifications, user stories or requirements,

which often is ok, but lacks aggression− no combinations, no unexpected situations, lame and boring− such tests have a hard time finding (interesting) bugs

� Better automation !− when unneeded details are left out of tests, they don't have to be

maintained− avoiding "over checking": creating checks that are not in the scope of

a test, but may fail after system changes− limit the impact of system changes on tests, making such impact

more manageable

I have become to believe that successful automation is less of a technical challenge as it is a test design challenge.

4/10/2014

17


Case for organizing tests in BIG projects

� Can help keep the volume down

� Isolate the complexities

� Efficient and re-usable automation

� Deal with changing requirements

� For example: much of tested subject matter is not system specific, but business specific− a mortgage is a mortgage


The Three “Holy Grails” of Test Design

� Metaphor to depict three main steps in test design

� Using "grail" to illustrate that there is no one perfect solution, but that it matters to pay attention (to search)

� About quality of tests, but most of all about scalability and maintainability in BIG projects

Right approach for each test module

Proper level of detail in the test specification

Organization of tests into test modules

4/10/2014

18


What's the trick...


What's the trick...

� Have or acquire facilities to store and organize your content

� Select your stuff

� Decide where to put what− assign and label the shelves

� Put it there

� If the organization is not sufficient anymore, add to it or change it

4/10/2014

19


Properties of a good Breakdown

� Test modules are well differentiated and clear in scope

� Reflects the level of tests

� Balanced in size and amount

� Modules are mutually independent

� Fit the priorities and planning of the project


Breakdown Criteria

� Straightforward Criteria− Functionality (customers, finances, management information, UI, ...)− Architecture of the system under test (client, server, protocol, sub

systems, components, modules, ...)− Kind of test (navigation flow, negative tests, response time, ...)

� Additional Criteria− Stakeholders (like "Accounting", "Compliance", "HR", ...)− Complexity of the test (put complex tests in separate modules)− Technical aspects of execution (special hardware, multi-station, ...) − Overall project planning (availability of information, timelines, sprints, ...)− Risks involved (extra test modules for high risk areas)− Ambition level (smoke test, regression, aggressive, O)

4/10/2014

20


What is probably not a good design

� Navigational and functional tests are mixed− for example "over checking": a test of a premium calculation also

checks the existence of a window

� You have to change all of them for every new release of the system under test

� All test modules have a similar design

� Test modules are dependent on each other

� You can’t start developing any test modules early in the life cycle


Symptoms

� Tediousness in the test and test automation process

� No sense of control

� Complaining people

� Unnecessary high test maintenance− changes in the system under test impact many tests− hard to understand which tests need to be modified

� Difficulties in running any test− teams start "debugging" tests

4/10/2014

21


Breakdown examples

� CRUD tests (Create, Read, Update, Delete) for all entity types in the app− like "order", "customer", "well", etc− for all: various types and situations

� Forms, value entry− does each form work (try to test form by form, not entity by entity)− mandatory and optional fields, valid and invalid values, etc− UI elements and their properties and contents− function keys, tab keys, special keys, etc

� Screen and transaction flows− like cancel an order, menu navigation, use a browser back and forward buttons, etc− is the data in the database correct after each flow

� Business transactions, business rules− identify situations that the tests need to try

� Function tests, do individual functions work− can I count orders, can I calculate a discount, etc

� End-to-end tests− like enter sale order, then check inventory and accounting

� Tests with specific automation needs− like multi station tests

� Tests of non-UI functions

� High ambition tests (aggressive tests)− can I break the system under test


Identifying the modules

Step 1: top down: establish main structure:

� analyze what the business is and what the system does?

� how is it technically organized?

� do other “primary criteria” apply?

� use the list in the "breakdown examples" slide as a starting point− see it as "low hanging fruit": items that tend to apply well in many projects

� also visit the “secondary criteria”− not always applicable, but can help to refine the design further

Step 2: bottom up: refine, complete:

� study individual functionalities and checks (like from exist test cases)

� and identify test modules for them if needed

� identify and discuss any additional criteria and needed testing situations

� review and discuss the resulting list(s) of test modules

� create some early drafts of test modules and adjust the list if needed

Repeat steps 1 and 2 if needed.

4/10/2014

22

2009 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.© 2011 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.

Test Module Scope Prio Status

Model Life CyclesCreate, store, delete Models (=formula + data), as part of SYS

sessions1 pass

Result Life Cycles Create, store outputs. See them in the process store. 1 pass

Formula Life Cycles Create, edit, manage, remove formulas 2 pass

Formula Editor buttons, operations, undo 3 pass

Repository

display of the Modeler repository, presence of user formulas,

drag and drop usage. Effect of changing repository folder

(environment variable)

1 failed

Model Store in Repository presence, re-run, delete 1 pass

Repository UI example: selecting an item shows its description 2 errors

Formula EvaluationCorrectness of results, valid/invalid arguments, boundary

analyses, special arguments1 pass

Built-in FormulasPresence, correctness, valid/invalid arguments, boundaries,

special arguments, equivalence classes1 pass

Data Table AssociationAssociate tabels view, change and remove associations, data

applicability, for existing and defined formulas2 pass

Quick Access buttonsLife cycle of Quick Access buttons, correctnes for the built-in

ones3 dev

Formula argumentspresence, argument types, argument entry, parameters,

defaults2 pass

arguments for Built-in

Formulas

arguments, argument types and defaults for each pre-defined

formula2 failed

Area Of Interest Relations defaulting, tree visibility, select/deselect, …. 1 pass

Model ExecutionModel times, start, stop (cancel), restart ("chunks",

"timeboxes", ... needs more information)3 pass

Graphics graphical representation of various data types and data sets 1 pass

Graphics Viewing zoom, select, drag and drop (no 3d now) 1 pass

Administration users, projects, authorization 1 pass

Model results in central

database

storing, removing, using, correctness, … (there are some other

applications, mostly legacy, that can do the same Models to

compare)

1 pass

Modeler UI various controls, panels, tabs 2 pass




sessions1 pass




Repository




1 failed










ones3 dev


defaults2 pass


Formulas


formula2 failed








database



compare)

1 pass


4/10/2014

23




sessions1 pass




Repository




1 failed










ones3 dev


defaults2 pass


Formulas


formula2 failed








database



compare)

1 pass





sessions1 pass




Repository




1 failed










ones3 dev


defaults2 pass


Formulas


formula2 failed








database



compare)

1 pass


4/10/2014

24




sessions1 pass




Repository




1 failed










ones3 dev


defaults2 pass


Formulas


formula2 failed








database



compare)

1 pass



Test Module Scope Prio Build 1 Build 2 Build 3


sessions1 pass pass pass

Result Life Cycles Create, store outputs. See them in the process store. 1 pass pass pass

Formula Life Cycles Create, edit, manage, remove formulas 2 pass pass pass

Formula Editor buttons, operations, undo 3 pass pass pass

Repository




1 failed failed failed

Model Store in Repository presence, re-run, delete 1 pass pass pass

Repository UI example: selecting an item shows its description 2 errors errors errors


analyses, special arguments1 pass pass pass


special arguments, equivalence classes1 pass pass pass


applicability, for existing and defined formulas2 pass pass pass


ones3 dev dev dev


defaults2 pass pass pass


Formulas


formula2 failed failed failed

Area Of Interest Relations defaulting, tree visibility, select/deselect, …. 1 pass pass pass


"timeboxes", ... needs more information)3 pass pass pass

Graphics graphical representation of various data types and data sets 1 pass pass pass

Graphics Viewing zoom, select, drag and drop (no 3d now) 1 pass pass pass

Administration users, projects, authorization 1 pass pass pass


database



compare)

1 pass pass pass

Modeler UI various controls, panels, tabs 2 pass pass pass

4/10/2014

25


Questions for Test Design

� Does your organization make something like a high level test design?

� If yes, how do you document it?


Case Study

� Large IT provider

� New version of one of their major web-sites

� Test scope was user acceptance test (functional acceptance)− the users were the “business owners”

� Development was off-shore

4/10/2014

26


Case Study

� Test development was done separate from automation− time-line for test development: May – Oct− time-line for automation (roughly): Jan – Feb

� All tests were reviewed and approved by the business owners− acceptance was finished by the end of the test

development cycle


Example of a Test Development Plan

Nr Module Business Owner Date to BO

1 Portal Navigation, Audience Robyn Peterson 05 / 232 Portal Navigation, Search Ted Jones 05 / 273 Membership, registration Steve Shao 06 / 034 Portal Navigation, Category Ted Jones 06 / 085 Portal Navigation, Topic and Expert Ted Jones 06 / 136 Access Control Mike Soderfeldt 06 / 177 Portal Navigation, Task Ted Jones 06 / 228 Contact DSPP Ted Jones 06 / 279 Portal search Mike Soderfeldt 07 / 0110 Membership, review and update Steve Shao 07 / 0511 Program contact assignment Alan Lai 07 / 1112 Company, registration Steve Shao 07 / 1413 Catalog, view and query Robyn Peterson 07 / 1914 Site map Ted Jones 07 / 2515 Membership, affiliation Steve Shao 07 / 2816 Learn about DSPP Ted Jones 08 / 0117 Products and services Steve Shao, Robyn Peterson 08 / 0818 What's new Ted Jones 08 / 1119 Company, life cycle Steve Shao, Alan Lai 08 / 1720 Specialized programs Ted Jones, Steve Shao 08 / 2221 Customer surveys Ted Jones 08 / 2922 Software downloads Mike Soderfeldt 09 / 0123 Newsletters Ted Jones 09 / 0624 Internationalization and localization Ted Jones 09 / 1325 Membership, life cycles Steve Shao 09 / 1926 Collaboration, forums Ted Jones 09 / 2327 Collaboration, blogs Mike Soderfeldt 09 / 2828 Collaboration, mailing lists Ted Jones 10 / 03

4/10/2014

27


Review Process with Stake Holders

Test Team sends draft Module to Stake Holder

Stake Holder reviews: - coverage

- correctness

Stake Holder returns notes:

- additions- corrections

Test Team receives and processes notes

changes needed?Stake Holder returns

notice of approval

Test Team marks the Module as "Final"

END

no

yes

START


Case Study, Results

� All tests were developed and reviewed on schedule− many notes and questions during test development phase

� The automation was 100% of the tests− all actions were automated, thus automating all test modules

� The test development took an estimated 18 person months− one on-shore resource, two off-shore resources

� The automation took between one and two months− focused on actions− most time was spent in handling changes in the interface (layout of pages

etc)

4/10/2014

28


Case: The French Director

� Mid size company

� Struggling under high pressure

� Testing of their main product, standard financial software

� Control and priority main issue

� Unfamiliar business culture

� Main instrument: module break down


Test Modules versus Test Cases

� The test module is a bigger unit in the test design− easier to identify− a chapter rather than a paragraph− easier to plan and manage, as a product (can be treated as part of product

backlog in scrum projects)

� Better flow of execution− each test case can set up for the next one− keep test modules independent, test cases can be dependent

� Test cases become creative output, rather than stifling input− avoids having to define all test cases at once early in the process

� Clear scope helps to identify cases, actions and checks− using "test objectives" to further detail scope− had a significant effect on maintainability

4/10/2014

29


"Thou Shall Not Debug Tests..."

� Large and complex test projects can be hard to "get to run"

� If they are however, start with taking a good look again at your test design...

� Rule of thumb: don't debug tests. If tests don't run smoothly, make sure:− lower level tests have been successfully executed first -> UI flow in the AUT

is stable− actions and interface definitions have been tested sufficiently with their own

test modules -> automation can be trusted− are you test modules not too long and complex?


What about existing tests?

� Compare to moving house:− some effort can't be avoided− be selective, edit your stuff,

• look at the future, not the past

− first decide where to put what, then put it there− moving is an opportunity, you may not get such chance again soon

� Follow the module approach− define the modules and their scope as if from scratch− use the existing test cases in two ways:

• verify completeness• harvest and re-use them for tests and for actions

− avoid porting over "step by step", in particular avoid over-checking

4/10/2014

30


Defining test runs using "test suites"

Build Acceptance Test

Smoke Test

System Test

Functional Acceptance Test

Integration Test


Grail 2: Approach per Test Module

� Plan the test module:− when to develop: is enough specification available− when to execute: make sure the functionality at action level is well-

tested and working already

� Process:− do an intake: understand what is needed and devise an approach− analyze of requirements− formulate "test objectives"− create "test cases"

� Identify stakeholders and their involvement:− users, subject matter experts− developers− auditors

� Choose testing techniques if applicable:− boundary analysis, decision tables, etc

4/10/2014

31


Eye on the ball, Scope

� Always know the scope of the test module

� The scope should be unambiguous

� The scope determines many things:− what the test objectives are− which test cases to expect− what level of actions to use− what the checks are about and which events should

generate a warning or error (if a “lower” functionality is wrong)


What I have seen not work

� Lots of detailed steps when those steps are not the focus of the test

� All actions high level, or all actions low level

� "Over-Checking": having checks that don't fit the scope of the test

� Over-use of data externalization (data driven) with values coming from files or data sets without a clear testing reason

� Combinatorial explosions: test all ... for all ... in all ...

� Many tests for forms and dialogs, little tests for business processes

4/10/2014

32

2009 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.

An example test for the Modeler

model name arguments formula

create model vegas winner x 10*x

argument value

set argument x some money

model name expected

run model vegas winner a lot more money

2009 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.© 2014 LogiGear Corporation. All rights reserved.

Too detailed?

Step Name Description Expected

step 16 Click the new formula button to start a new

calculation.

The current formula is cleared. If it had not

been save a message will show

step 17 Enter "vegas winner" in the name field The title will show "vegas winner"

step 18 Open the formula editor by clicking the '+'

button for the panel "formula editor"

The formula editor will show with an empty

formula (only comment lines)

step 19 Add some lines and enter "10*x;" The status bard will show "valid formula".

There is a "*" marker in the title

step 20 Click the Save formula button The formula is saved, the "*" will disappear

from the title

step 21 Open the panel with the arguments by

clicking the '+' button

There two lines, for 'x' and 'y'

step 22 Click on the value type cell and select

"currency"

A button to select a currency appears, with

default USD

step 23 Click on the specify argument values link The argument specification dialog is shown

4/10/2014

33


Detail out the scope with test objectives

...TO-3.51 The exit date must be after the entry date...

test objective TO-3.51

name entry date exit date

enter employment Bill Goodfellow 2002-10-02 2002-10-01check error message The exit date must be after the entry date.


Examples of Testing Techniques

� Equivalence class partitioning− any age between 18 and 65

� Boundary condition analysis− try 17, 18, 19 and 64, 65, 66

� Error guessing− try Cécile Schäfer to test sorting of

a name list

� Exploratory− "Exploratory testing is

simultaneous learning, test design, and test execution", James Bach, www.satisfice.com

� Error seeding− deliberately injecting faults in a test

version of the system, to see if the tests catch them

− handle with care, don't let the bugs get into the production version

� Decision tables− define possible situations and the

expected responses of the system under test

� State transition diagrams− identify "states" of the system, and

have your tests go through each transition between states at least once

� Jungle Testing− focus on unexpected situations,

like hacking attacks

� Soap Opera Testing− describe typical situations and

scenarios in the style of episodes of a soap opera, with fixed characters

− high density of events, exaggerated

− make sure the system under test can still handle these

4/10/2014

34


"Jungle Testing"

� Expect the unexpected− unexpected requests− unexpected situations (often data oriented)− deliberate attacks− how does a generic design respond to a specific unexpected event?

� Difference in thinking− coding bug: implementation is different from what was intended/specified− jungle bug: system does not respond well to an unexpected situation

� To address− study the matter (common hack attacks, ...)− make a risk analysis− make time to discuss about it (analysis, brainstorm)− involve people who can know− use "exploratory testing" (see James Bach's work on this)− use an agile approach for test development− consider randomized testing, like "monkey" testing

New York. The city of a million stories. Half of them aretrue, the other half just haven't happened yet -- Dr Who


Soap Opera Testing

� Informal scenario technique to invite subject-matter experiences into the tests, and efficiently address multiple objectives

� Using a recurring theme, with “episodes”

� About “real life”

� But condensed

� And more extreme

� Typically created with a high involvement of end-users and/or subject-matter experts

� It can help create a lot of tests quickly, and in an agile way

4/10/2014

35


Lisa Crispin: Disorder Depot . . .

There are 20 preorders for George W. Bush action figures in "Enterprise", the ERP system, awaiting the receipt of the items in the warehouse.

Finally, the great day arrives, and Jane at the warehouse receives 100 of the action figures as available inventory against the purchase order. She updates the item record in Enterprise to show it is no longer a preorder.

Some time passes, during which the Enterprise background workflow to release preorders runs. The 20 orders are pick-released and sent down to the warehouse.

Source: Hans Buwalda, Soap Opera Testing (article), Better Software Magazine, February 2005



Then Joe loses control of his forklift and accidentally drives it into the shelf containing the Bush action figures. All appear to be shredded to bits. Jane, horrified, removes all 100 items from available inventory with a miscellaneous issue. Meanwhile, more orders for this very popular item have come in to Enterprise.

Sorting through the rubble, Jane and Joe find that 14 of the action figures have survived intact in their boxes. Jane adds them back into available inventory with a miscellaneous receipt.

4/10/2014

36



This scenario tests

• Preorder process

• PO receipt process

• Miscellaneous receipt and issue

• Backorder process

• Pick-release process

• Preorder release process

• Warehouse cancels


Vary your tests?

� Automated tests have a tendency to be rigid, and predictable

� Real-world situations are not necessarily predictable

� Whenever possible try to vary:− with select other data cases that still fit the goal of tests− with randomized behavior of the test

4/10/2014

37


Generation and randomization techniques

� Model-based− use models of the system under test to create tests− see: Harry Robinson, www.model-based-testing.org, and Hans Buwalda, Better

Software, March 2003

� Data driven testing− apply one test scenario to multiple data elements− either coming from a file or produce by an automation

� "Monkey testing"− use automation to generate random data or behavior− "smart monkeys" will follow typical user behavior, most helpful in efficiency− "dumb monkeys" are more purely random, may find more unexpected issues− long simulations can expose bugs traditional tests won't find

� Extended Random Regression− have a large database of tests− randomly select and run them, for a very long time− this will expose bugs otherwise hidden− see Cem Kaner e.a.: "High Volume Test Automation", StarEast 2004


Data Driven Testing

� Separate test logic from the data

� Possible origins for the data:− earlier steps in the test− data table− randomizer, or other formula− external sources, like a database query

� Use "variables" as placeholders in the test case, instead of hard values

� Data driven is powerful, but use modestly:− value cannot be known at test time, or changes over time− having many data variations is meaningful for the test

4/10/2014

38


Variables and expressions with keywords

� This test does not need an absolute number for the available cars, just wants to see if a stock is updated

� As a convention we denote an assignment with ">>"

� The "#" indicates an expression

TEST CASE TC 02 Rent some more cars

car available

get quantity Chevvy Volt >> volts

first name last name car

rent car John Doe Chevvy Volt

rent car John Doe Chevvy Volt

car expected

check quantity Chevvy Volt # volts - 2


Data driven testing with keywords

� The test lines will be repeated for each row in the data set

� The values represented by "car", "first" and "last" come from the selected row of the data set

TEST CASE TC 03 Check stocks

data set

use data set /cars

car available

get quantity # car >> quantity

first name last name car

rent car # first # last # car

car expected

check quantity # car # quantity - 1

repeat for data set

DATA SET cars

car first last

Chevvy Volt John Doe

Ford Escape Mary Kane

Chrysler 300 Jane Collins

Buick Verano Tom Anderson

BMW 750 Henry Smyth

Toyota Corolla Vivian Major

4/10/2014

39


Combinations

� Input values− determine equivalence classes of values for a variable or field− for each class pick a value (or randomize)

� Options, settings

� Configurations− operating systems, operating system versions and flavors

• Windows service packs, Linux distributions

− browsers, browser versions− protocol stacks (IPv4, IPv6, USB, ...)− processors− DBMS's

� Combinations of all of the above

� Trying all combinations will spin out of control quickly


Pairwise versus exhaustive testing

� Group values of variables in pairs (or tuples with more than 2)

� Each pair (tuple) should occur in the test at least once− maybe not in every run, but at least once before you assume "done"− consider to go through combinations round-robin, for example pick a different

combination every time you run a build acceptance test− in a NASA study:

• 67 percent of failures triggered by a single value• 93 percent by two-way combinations, and • 98 percent by three-way combinations

� Example, configurations− operating system: Windows XP,

Apple OS X, Red Hat Enterprise Linux− browser: Internet Explorer, Firefox, Chrome− processor: Intel, AMD− database: MySQL, Sybase, Oracle− 72 combinations possible, to test each pair: 10 tests

� Example of tools: − ACTS from NIST, PICT from Microsoft, AllPairs from James Bach (Perl)− for a longer list see: www.pairwise.org

� These techniques and tool are supportive only. Often priorities between platforms and values can drive more informed selection

Source: PRACTICAL COMBINATORIAL TESTING, D. Richard Kuhn, Raghu N. Kacker, Yu Lei, NIST Special Publication 800-142, October, 2010

4/10/2014

40


Grail 3: Specification Level, choosing actions

� Scope of the test determines the specification level

� As high level as appropriate, as little arguments as possible− Use default values for non-relevant arguments

� Clear names (usually verb + noun usually works well)− to standardize action names: standardize both the verbs and the nouns, so

"check customer" versus "verify client" (or vice versa)− tests are not C++ code: avoid "technical habits", like mixed case and (worse)

underlines

� Manage the Actions

� Document the Actions

� By-product of the test design


Case: American Bank

� Project for a new teller system

� Large, state of the art

� Many system releases, many adjustments

� Need for very high level of automation

� Over 1 million test lines, in over 650 test modules

� Initially little attention paid to "holy grails"− UI and functional tests in the same modules− virtually un-maintainable, came close to killing the project− test design forced upon the team by a powerful stakeholder

who did not care much for methods...

� Emergency re-organization of the test modules− after system changes the tests would run again within a day

4/10/2014

41


Example of using actions

In this real world example the first "sequence number" for teller transactions for a given day is retrieved, using a search function• the "#" means an expression, in this case a variable with todays business date• the ">>" means: assign to a variable for use later on in the test

key

type key {F7}

type key 3

page tab

locate page tab Scan Criteria

text

check breadcrumb general functions > search


select search scan direction Backward


enter value search business date match # bus date

source control

click search go

window control variable

get search results sequence number >> seq num

. . .


variable

get sequence number >> seq num

Example of using actions

In this real world example the first "sequence number" for teller transactions for a given day is retrieved, using a search function• the "#" means an expression, in this case a variable with todays business date• the ">>" means: assign to a variable for use later on in the test

4/10/2014

42


Low-level, high-level, mid-level actions

� Low-level: detailed interaction with the UI (or API)− generic, do not show any functional or business logic− examples: "click", "expand tree node", "select menu"

� High-level: represent a business function specific to the scope of the test− hide the interaction− examples: "enter customer", "rent car", "check balance"

� Mid-level: auxiliary actions that represent common sequences of low level actions− usually to wrap a form or dialog− greatly enhance maintainability− example: "enter address fields"

enter customer

enter address fields

enter select set . . .. . .


Identifying controls

� Identify windows and controls, and assign names to them

� These names encapsulate the properties that the tool can use to identify the windows and controls when executing the tests

4/10/2014

43


Mapping the interface

� An interface mapping (common in test tools) will map windows and controls to names

� When the interface of an application changes, you only have to update this in one place

� The interface mapping is a key step in your automation success, allocate time to design it well

INTERFACE ENTITY library

interface entity setting title {.*Music Library}

ta name ta class label

interface element title text Title:

interface element artist text Artist:

interface element file size text File size (Kb):

ta name ta class position

interface element playing time text textbox 4

interface element file type text textbox 5

interface element bitrate text textbox 6

ta name ta class position

interface element music treeview treeview 1


Some Tips to Get Stable Automation

� Make the system under test automation-friendly− consider this a key requirement ("must have")− screen elements have stable identifying properties− white box access to data, statuses, conditions, etc

� Use "active" timing, don't use hard coded waits

� Test the automation, in particular complex actions− before running tests with them

� Keep an eye on the test design

4/10/2014

44


� Look for properties a human user can't see, but a test tool can

� This approach can lead to speedier and more stable automation− interface mapping is often bottleneck, and source of maintenance problems− with predefined identifying property values in interface map can be created without "spy" tools− not sensitive to changes in the system under test− not sensitive to languages and localizations

� Examples:− "id" attribute for HTML elements− "name" field for Java controls− "AccessibleName" or "Automation ID" properties in .Net controls (see below)

Automation-friendly design: hidden properties


Mapping the interface using hidden identifiers

� Instead of positions or language dependent labels, an internal property "automation id" has been used

� The interface definition will be less dependent on modifications in the UI of the application under test

� If the information can be agreed upon with the developers, for example in an agile team, it can be entered (or pasted) manually and early on

INTERFACE ENTITY library

interface entity setting automation id MusicLibraryWindow

ta name ta class automation id

interface element title text TitleTextBox

interface element artist text SongArtistTextBox

interface element file size text SizeTextBox

interface element playing time text TimeTextBox

interface element file type text TypeTextBox

interface element bitrate text BitrateTextBox

ta name ta class automation id

interface element music treeview MusicTreeView

4/10/2014

45


Active Timing

� Passive timing− wait a set amount of time− in large scale testing, try to avoid passive timing altogether:

• if wait too short, test will be interrupted• if wait too long, time is wasted

� Active timing− wait for a measurable event− usually the wait is up to a, generous, maximum time− common example: wait for a window or control to appear (usually the test tool will do

this for you)

� Even if not obvious, find something to wait for...

� Involve developers if needed− relatively easy in an agile team, but also in traditional projects, give this priority

� If using a waiting loop− make sure to use a "sleep" function in each cycle that frees up the processor (giving the

AUT time to respond)− wait for an end time, rather then a set amount of cycles


Things to wait for...

� Wait for a last control or elements to load− developers can help knowing which one that is

� Non-UI criteria− API function− existence of a file

� Criteria added in development specifically for this purpose, like:− "disabling" big slow controls (like lists or trees) until they're done loading− API functions or UI window or control properties

� Use a "delta" approach:− every wait cycle, test if there was a change; if no change, assume that the

loading time is over:− examples of changes:

• the controls on a window• count of items in a list• size a file (like a log file)

4/10/2014

46


Alternatives to UI automation ("non-GUI")

� A GUI (Graphical User Interface) is only one example of an interface for interaction with a system under test

� Examples− HTTP and XML based interfaces, like REST− application programming interfaces (API’s)− embedded software− protocols− files, batches− databases− command line interfaces (CLI’s)− multi-media− mobile devices

� In many cases non-GUI automation is used since there simply is not GUI, but it can also often speed things up:− tends to be more straightforward technically, little effort needed to build up or maintain− once it works, it tends to work much faster and more stably than GUI automation

� In BIG testing projects routinely:− identify which non-GUI alternatives are available− as part of test planning: identify which tests qualify for non-GUI automation


Tools that can help manage BIG projects

� Application Lifecycle Management (ALM)− abundant now, mainly on the wings of agile− very good for control, team cooperation, and traceability− often relate to IDE's (like Microsoft TFS and Visual Studio)− examples: Rally, Jira, TFS

� Test Managers− as separate tools on their way out− morphing into or replaced by ALM options− examples: HP Quality Center, Microsoft Test Manager

� Test development and automation tools− develop and/or automate tests

• these are not the same, automation tools are not always so good for test development− examples are HP Quick Test Pro, Borland Silk, Selenium, FitNesse, Microsoft Coded UI, and LogiGear's

TestArchitect and TestArchitect for Visual Studio (our own products)

� Build tools− succeed the traditional "make" tools− in particular "continuous build" tools combine "make" functionality with source control systems to rebuild

components that have changed, either continuously or on set times, like nightly− can very well also run related tests (unit and functional), and act on the results (stop build, report, etc)− examples: Hudson, Jenkins, TFS

� Bug trackers− not only register issues, but also facilitate their follow up, with workflow features− often also part of other tools, and tend to get absorbed now by the ALMs− Examples: BugZilla, Mantis, Trac

4/10/2014

47


Tooling and Traceability

Reference item(ALM item, req, code module, ...)

Test Objective Test Case Execution Result

Test Module

Bug, issue

ALM, IDE,Project Mgr,

Req Mgr

Test Development ToolAutomation Tool

Execution ManagerContinuous Build Tool

Lab manager

Issue TrackerALM

Testing

Trace back


Test Execution

� Have an explicit approach for when and how to execute which tests

� Having a good high level test design will help to organize this

� Execution can be selective or integral− unit tests are typically executed selectively, possibly automatically

based on code changes in a system like SVN or TFS− for functional tests, decisions are needed:

• selective execution will be quicker and more efficient• integral execution may catch more issues ("bonus bugs")• generally extensive functional test execution will be related to releases, rather

than code check ins

− the ability to run "big testing" efficiently may determine how much can be done

4/10/2014

48


Environments, configurations

� Many factors can influence details of automation− language, localization− hardware− version of the system under test− system components, like OS or browser

� Test design can reflect these− certain test modules are more general− others are specific, for example for a language

� But for tests that do not care about the differences, the automation just needs to "deal" with them− shield them from the tests


Capture variations of the system under test in the actions and interface definitions, rather than in the tests (unless relevant there).Can be a feature in a test playback tool, or something you do with a global variable or setting.

Variation Variation Variation

"Variations"

"Master Switch"

Actions, Interface Definitions

. . .

4/10/2014

49


Possible set up of variations

linked variation

keyworded variation

Specify for example in a dialog when you start an execution:


Test Environments

� Physical• hardware• infrastructure• location• . . .

� Software• programs• data models• protocols• . . .

� Data• initial data• parameters / tables• . . .

• costs money• can be scarce

• configurations

• availability • manageability

4/10/2014

50


Dealing with data

� Constructed data is easier to manage− can use automation to generate it, and to enter it in the environment− result of test analysis and design, reflecting "interesting" situations− however, less "surprises": real life situations which were not foreseen

� Real-world data is challenging to organize− make it a project, or task, in itself− make absolutely sure to deal with privacy, security and legal aspects

appropriately• study this, ask advice• apply appropriate "scrubbing"

� Consider using automation to select data for a test− set criteria ("need a male older than 50, married, living in Denver"),

query for matching cases, and select one randomly (if possible a different one each run)

− this approach will introduce variation and unexpectedness, making automated tests stronger and more interesting


Unattended testing...

� When a test cannot pass, it can be:− a difference between expected and recorded values or behavior, as a result

of a check designed by the tester: this is a fail− the automation encounters a problem, like a window or control doesn't show,

that is not part of a check: this is an error

� An error can disrupt the test flow, and you may want to catch and handle it properly:− skip smaller or larger parts of the ongoing test− bring the system back in a known state (typically: close any open windows,

go to the main screen)− make sure the report clearly indicates these kind of problems, to avoid false

positives− example "on error action" that executes a predefined action that will do

recovery

� However, better is to avoid these situations− lots of efforts needed for unattended testing should raise questions about test

design or quality of the automation

4/10/2014

51


"Known bug" problem

� Not uncommon in large scale systems− typically related to a version of the system under test

� A known bug may:− generate fails you want to ignore, also in statistics− throw off automation

� One possible approach, a "known bug" marker, so you filter out "new" issues

� If many known-bug situations occur, take another look at your high level test design


Virtualization

� Virtual machines rather than physical machines− allow "guest" systems to operate on a "host" system− host can be Windows, Linux, etc, but also a specialized "hypervisor"− the hypervisor can be "hosted" or "bare metal"

� Main providers:− VMWare: ESX and ESXi− Microsoft: Hyper-V− Oracle/Sun: Virtual Box− Citrix: Xen (open source)

� Hardware support gets common now− processor, chipset, i/o− Like Intel's i7/Xeon

� For most testing purposes you need virtual clients, not virtual servers− most offerings in the market currently target virtual servers, particularly data centers

� Virtual clients will become more mainstream with the coming of VM's as part of regular operating systems− Windows 8: Hyper-V− Linux: KVM

4/10/2014

52


Virtualization, a testers dream...

� In particular for functional testing

� Much easier to define and create needed configurations− you basically just need storage− managing this is your next challenge

� One stored configuration can be re-used over and over again

� The VM can always start "fresh", in particular with − fresh base data (either server or client)− specified state, for example to repeat a particular problematic automation

situation

� Can take "snap shots" of situations, for analysis of problems

� Can use automation itself to select and start/stop suitable VM's− for example using actions for this− or letting an overnight or continuous build take care of this


Virtualization, bad dream?

� Performance, response times, capacities

� Virtual machine latency can add timing problems− see next slide− can be derailing in big test runs

� Management of images− images can be large, and difficult to store and move around

• there can be many, with numbers growing combinatorial style• configuration in the VM can have an impact, like fixed/growing virtual disks

− distinguish between managed configurations and sandboxes− define ownership, organize it− IT may be the one giving out (running) VM's, restricting your flexibility

� Managing running tests in virtual machines can take additional efforts on top of managing the VM's themselves− with the luxury of having VM's the number of executing machines can

increase rapidly− one approach: let longer running tests report their progress to a central

monitoring service (various tools have features for this)

4/10/2014

53


Virtualization: "time is relative"

� Consider this waiting time loop, typical for a test script:− endTime = currentTime + maxWait− while not endTime, wait in 100 millisecond intervals

� When the physical machine overloads VM's can get slow or have drop outs, and endTime may pass not due to AUT latency− GetLocalTime will suffer from the latency− GetTickCount is probably better, but known for being unreliable on VM's

� Therefore tests that run smooth on physical machines, may not consistently do so on VM's. The timing problems are not easy to predict

� Possible approaches:− in general: be generous with maximum wait times if you can− don't put too many virtual machines on a physical box− consider a compensation algorithm, for example using both tick count and clock time


Virtual machines, capacity

� Key to pricing is number of VM's that can run in parallel on a physical machine

� An automated test execution will typically keep a VM more busy than human use

� Factors in determining VM/PM ratio:− memory, for guest OS, AUT, test tooling− storage devices (physical devices, not disk images)− processors, processor cores− specific hardware support (becoming more common)

• processor, chipset, I/O

We started regression with 140 VMs.Very slow performance of

Citrix VM clients.

4/10/2014

54


Building up virtualization

� Pay attention to pricing:− beefed up hardware can increase VM's/box ratio, but at a price− software can be expensive depending on features, that you may not

need

� In a large organization, virtual machines are probably available− make sure to allocate timely (which can be long before you get there

with your sprints)− keep in mind the capacity requirements

� Logical and physical management− which images, the wealth of possible images can quickly become

hard to see forest through the trees− physical management of infrastructure is beyond this tutorial

� Minimum requirement: snapshots/images− freeware versions don't always carry this feature− allow to set up: OS, environment, AUT, tooling, but also: data, states


Infrastructure

� For large scale test execution this needs attention− physical infrastructure, but also how to use it

� Also consider managing infrastructure and test execution as a separate task− in or out of the team− avoid slowing down development (of system, test and/or

automation)

4/10/2014

55


Remote execution, servers

� Allowing execution separately from the machines the testers and automation engineers are working on increases scalability

� Large scale text execution, in particular with VM's, like to have:− lots of processing power, lots of cores− lots of memory

� Test execution tends to care less about:− storage− networking

� Test execution facilities tend to be a bottle neck very quickly in big testing projects− the teams can use whatever they can get

� First step up: give team members a second machine

� Second step up: use servers, users coordinate their use of them

� Third step up: major infrastructures with organized allocation


Tower Servers

� Smaller shops (smaller companies, departments)

� Affordable, simple, first step up from clients execution

� Not very scalable when the projects get larger

4/10/2014

56


Rack Servers

� Well scalable

� Pricing not unlike tower servers

� Tend to need more mature IT expertise


Server Blades

� Big league infrastructure, high density, very scalable

� Tends to be pricey, use when space and energy matters

� Usually out of sight for you and your team

4/10/2014

57


Cloud

� Cloud can be target of testing− normal tests, plus cloud specific tests

• functional, load, response times

− from multiple locations− moving production through data centers

� Cloud can be host of test execution− considerations can be economical or organizational− providers offer imaging facilities, similar to virtual machines− make sure machines are rented and returned efficiently

� Public cloud providers like EC2 offer API's, so your automation can automatically allocate and release them− be careful, software bugs can have costing consequences− for example, consider having a second automation process to double-check cloud

machines have been released after a set time

� Note: public cloud is not taking of as fast as expected, cloud services, and private clouds, taking of much faster

(Xinhua Photo)


Cloud Providers

Source: Jack of All Clouds, January 2011http://www.jackofallclouds.com/2011/01/state-of-the-cloud-january-201/

4/10/2014

58


Cloud growth

� Growth of public clouds not as big as expected

� Cost benefits not necessarily convincing− low startup cost, but long ongoing cost

� See also: news.cnet.com/8301-13556_3-20063361-61.html

source: IDC forecast, 2010


Cloud, example pricing, hourly rates

� Source: Amazon EC2 (my interpretation, actual prices may vary)

Linux Windows

Small 0.085 0.12 1.7 GB, 1 core (32 bits)

Large 0.34 0.48 7.5 GB, 4 cores

Extra Large 0.68 0.96 15 GB, 8 cores

High memory

Extra Large 0.50 0.62 17.1 GB, 6.5 core

Double Extra Large 1.00 1.24 34.2 GB, 13 cores

Quadruple Extra Large 2.00 2.48 68.4 GB, 26 cores

High CPU

Medium 0.17 0.29 1.7 GB, 5 core (32 bits)

Extra Large 0.68 1.16 7 GB, 20 cores

4/10/2014

59


Cloud, example economy

� Not counting possible use of VM's within the buy option

� Also not counting: additional cost of ownership elements for owning or cloud (like IT management, contract and usage management)

� Impressions: − cloud could fit well for bursty testing needs, which is often the case− for full continuous, or very frequent, testing: consider buying− hybrid models may fit many big-testing situations: own a base capacity, rent

more during peak use periods

small large extra

Windows $0.12 $0.48 $0.96

buy (estimate) $300 $650 $900

hours to break even 2,500 1,354 938

months (24 / 7) 3.4 1.8 1.3


Data centers can go down

However, disruption could have been minimized by using multiple data centers

4/10/2014

60



This time, it did involve multiple data centers . . .



Service providers can occasionally go down too

4/10/2014

61


Cloud, usage for special testing needs

� Multi-region testing − Amazon for example has several regions

• US East, Northern Virginia• US West, Oregon, Northern California• EU, Ireland• Asia Pacific, Singapore, Tokyo• South America, Sao Paulo

− be careful that data transfers between regions costs money (0.01/GB)

� Load generation− example: "JMeter In The Cloud"

• based on the JMeter load test tool• uses Amazon AMI's for the slave machines• allows to distribute the AMI's in the different regions of Amazon• see more here:

aws.amazon.com/amis/jmeter-in-the-cloud-a-cloud-based-load-testing-environment


Questions for Infrastructure

� What kind of infrastructure does your organization use for testing?

� What is the role of virtualization, now or in the future?

� Are you using a private or a public cloud for testing?

4/10/2014

62


Example of a cloud system under test

� source: Windows Azure reference platform


Approaches

� Automation does not have to be black box− for very big systems, a separate black box automation effort may not

be efficient− and building and keeping lab situations might be cumbersome− some simple hooks can greatly help already− remember... this is about automation, not test design.

� Make testability part of requirements and architecture− a key question should not just be "how do I design this", but "how do I

test this" (test design, automation)− some cloud/web systems are changed frequently, and tested "live"

• "Testing in Production (TiP)"

− allow redirection of some or all traffic through another version of a component or layer

� Example: reverse proxy's enabling A/B testing

see also: Ken Johnston's chapter in the book of Dorothy Graham and Mark Fewster, and his keynote at StarWest 2012

4/10/2014

63


A/B testing with a reverse proxy

� Watch your test design, easy to drown in technical solutions only

� B could be a real-life user or also a keyword driven test machine

� A/B testing means part of traffic is routed through a different server or component (see if it works, and/or how users react)

� A similar strategy could be done at any component level

A

A

B

ReverseProxy

UsersServers

A

B

newcurrent

AB


Organization

� Much of the success is gained or lost in how you organize the process− part of the teams− who does test design− who does automation− what to outsource, what to keep in-house

� Write a plan of approach for the test development and automation− scope, assumptions, risks, planning− methods, best practices− tools, technologies, architecture− stake holders, including roles and processes for input and approvals− team− . . .

� Assemble the right resources− testers, lead testers− automation engineer(s)− managers, ambassadors, ...

Test design is a skill . . .Automation is a skill . . . Management is a skill . . .

. . . and those skills are different . . .

4/10/2014

64


Industrial Organization

� Large scale testing can move from a "design" to a "production" focus− mostly applies to test execution, but also seen for test development− this not black and white, both paradigms can occur in the same projects− this is often more easy to outsource than development

� A production organization is different a development organization− this is not unique for software− different professional culture− emphasis more on delivery and scale, "thinking big"− discipline rather than creativity, "get stuff done"− activities are like planning, control, logistics, information


Task in "production" (test execution)

� Keeping the tests running

� Allocating resources

� Respond to hick-ups

� Analyze and address automation issues

� Address fails or other testing outcomes− including dealing with "known bugs"− part of a bigger team

4/10/2014

65


Stake Holders

Test Development

Test Automation

Technology/Infrastructure

ProductionMarketing/Sales

System Development

End User Departments

Quality Assurance

Management

After Sales/Help Desk

Customers

Vendors

GovernmentAgencies

Publicity

EXTERNAL INTERNAL


Team roles, examples

� Test development

� Automation

� Planning and managing the test runs

� Managing environments

� Managing infrastructure

� Dealing with stakeholders

� Analysis of results, and follow up

� Reporting

4/10/2014

66


Test Development and Automation in sprints

Test ModuleDefinition(optional)

Test Module Development

Interface Definition

Action Automation

Test Execution

Sprint ProductsProduct Backlog

Test re-use

Automation re-use

product owner

teamprod owner & team

User storiesDocumentation

Domain understanding

Acceptance CriteriaPO Questions

SituationsRelations

Agile life cycle

Test development

Main Level Test Modules

Interaction Test Modules

Cross over Test Modules


Test automation in sprints

� Try keep the main test modules at a similar level as the user stories and acceptance criteria

� Aim for "sprint + zero", meaning: try to get test development and automation "done" in the same sprint, not the next one− next one means work clutters up, part of team is not working on the same sprint, work is

done double (manually and automated), ...

� Make sure you can do the interface mapping by hand (using developer provided identifications)− can do earlier, before UI is finalized, and − recording of actions (not tests) will go better

� Also plan for additional test modules:− low-level testing of the interaction with the system under test (like UI's)− crossing over to other parts of the system under test

� There should be agreement on the method(s) for testing and automation

� The team should include the skills and experienced needed for automated testing and the approach(es) taken for it

4/10/2014

67


Fitting in sprints

� Agree on the approach:− questions like does "done" include tests developed and automated?− do we see testing and automation as distinguishable tasks and

skillsets− is testability a requirement for the software

� Create good starting conditions for a sprint:− automation technology available (like hooks, calling functions, etc)− how to deal with data and environments− understanding of subject matter, testing, automation, etc

� Make testing and automation part of the evaluations

� Address tests and automation also in hardening sprints

� Just like for development, use discussions with the team and product owners to deepen understanding:− also to help identify negative, alternate and unexpected situations


Testing as a profession

� "Do thorough acceptance testing, but not only by the customer"− source: "Agile Software Testing in a Large-Scale Project", Israeli Air

Force

� Focus on tests, not development:− what can be consequences of situations and events− relieve developers

� Knowledge and experience with testing techniques and principles

� The challenge for the tester in the new era is to become a more credible professional tester, − not a pseudo programmer− part of the team

� Forcing a nontechnical tester to become a programmer may lose a good tester and gain a poor programmer

4/10/2014

68


Automation is a profession too

� Overlaps with regular system development, but not same

� Less concerned with complex code structures or algorithms

� More concerned with navigating through other software efficiently, dealing with control classes, obtaining information, timing, etc− if you would compare developers to "creators", automation engineers might

be likened to "adventurers"...

� The automation engineering role can also be a consultant:− for test developers: help express tests efficiently− for system developers: how to make a system more automation friendly − important player in innovation in the automated testing


Questions for Organization

� How is your testing currently organized (who is doing what)?− test design− test development− automation− execution− assessment of release readiness

� Do you use agile? If yes, is there a role for a test professional? And for an automation professional?

4/10/2014

69


Reporting

� Aim at needs:− avoid lengthy automated reports, have bottom line numbers− reports for stake holders− reporting for the team

� Reporting for a big testing project is about:− test and automation progress− production (running the tests)− results (aimed at system under test)

� Teams need (relevant) details− what happened, reproducibility, ...− either the tests, the automation, or the system under test− overall situations, with an ability to "drill down" to problem areas

� Management needs:− status, expectations, issues (realistic! bad news matter, you get punished for not telling)− bottom lines, plan versus reality confrontation− dates, efforts, used resources, costs, run times, ...− never allow planned numbers or dates to be "updated"

� Also for reporting, test organization is a key driver


War rooms

� Helpful if response times are critical, and a need for cooperation, towards the same goal− similar grounds as for agile scrum rooms

� Set up at critical times, like before important deadlines, or during critical releases

� Can temporarily bring together multiple parties, that normally are not co-workers− like competitor vendors

� Pay attention to physical conditions− machines, monitors, white boards, meeting places, headsets, ...− food, drinks, ...

� The test execution cycle should match the needs of the war room approach− fast turnarounds− effortless− completeness− selective or integral

See also: "Your Game is Live, Now What?", Jane Fraser, Electronic Arts

4/10/2014

70


Globalization....


Main Challenges

� Other countries

� Distances

� Time differences

4/10/2014

71


Globalization

� Three Challenges:− another countries, other cultures− geographic distances− time differences

� Seven "Patterns":− "Solution"− "Push Back"− "Time Pressure"− "Surprises"− "Ownership"− "Mythical Man Month"− "Cooperation"


Challenge: Other Country

4/10/2014

72


Other Country

� Differences in culture− more on the next slide...

� Different languages, and accents

� Differences in education− style, orientation and contents− position of critical thinking, factual knowledge, practice, theory,...− US, British, French, Asian, ...

� Differences in circumstances− demographics− economy, infrastructure− politics

� Apprehension on-shore and off-shore about job security doesn't help in projects− management responsibility: understand your strategic intentions, and their consequences, and clarify

them− be realistic in cost and benefit expectations


More on Culture...

� Regional culture. There are numerous factors:− very difficult to make general statements

• many anecdotes, stories and perceptions, some are very helpful, some have limited general value

• not sure on impact of regional culture (see also [Al-Ani])− numerous factors, like history, religion, political system

• e.g. valuing of: critical thinking, theory, bottom-line, relations, status, work-ethic, bad news, saying 'no'

• entertaining guests, eating habits, alcohol, meat, humor, etc• position of leaders, position of women managers• mistakes can be benign and funny, but also damaging, visibly or hidden, in particular perceived

disrespect hurts

� Organizational culture− can be different from country to country, sector to sector, company to company, group to group− I feel this to be at least as strong than regional culture (see for example [Al-Ani])− you can have at least some control over this

� Professional cultures− for example engineers, QA, managers, ...

� Some ideas to help:− get to know each other (it helps, see for example [Gotel])− study the matter, and make adaptations

4/10/2014

73



4/10/2014

74



4/10/2014

75


Challenge: Distance


Distance

� Continuous logistical challenges

� Lots of costs, and disruptions, for traveling

� Distance creates distrust and conflict− could be "normal" behavior, inherent to humans

� Complex coordination can create misunderstandings− on technical topics− on actions, priorities, and intentions

4/10/2014

76


Challenge: Time difference


Challenge: Time difference

� Additional complication for communication and coordination

� Places a major burden on both on-shore and off-shore staff− having to work evenings and/or early mornings− potential for exhaustion, lack of relaxation, mistakes, irritation

� Can easily lead to loss of time at critical moments

� Some solutions:− manage this actively− constantly seek to optimize task and responsibility allocation− build the on-shore and off-shore organizations to match− seek ways to save meeting time, like optimal information handling

4/10/2014

77


Effect of time difference

Test Module: “Segment Y, Default Settings”

Windows Linux

TestArchitect 5 ~ 4:16 m ~ 4:28 m

TestArchitect 6 ~ 11:00 m ~ 8:00 m

Report from the team to the US management . . .

Performance comparison TestArchitect 5 and 6


Patterns

� Experiences seem to follow patterns− at least our own experiences do− variations are numerous, but seem to follow similar lines− following are examples, not limitative

� It can help to recognize patterns quickly, and act upon them

� Resolutions have side-effects, can introduce new issues− for example strengthening local management means less direct

contact with the project members doing the work

� Just about every pattern occurs in every direction− from your perspective regarding "them"− their perspective on you, or each other− sometimes equaling, sometimes mirroring

4/10/2014

78


Pattern: "The Solution"

� Typical sequence of events:− the team finds a problem in running a test− the team discusses it and comes up with a "solution"− the solution: (1) creates issues, and (2) hides the real

problem

� Better way:− define as an issue− discuss with project manager and customer


Pattern: "Push Back"

� US side, or customer, gives bad direction

� Team doesn't like it, but feels obliged to follow orders

� The result is disappointing

� Team is blamed− and will speak up even less next time

� Better way:− discuss with the principal/customer at multiple levels

• strategic about direction, operational day-to-day

− empower and encourage the team to speak up− write plans of approach, and reports

4/10/2014

79


Pattern: "Time Pressure"

� Deadline must be met− no matter what− use over-time− "failure is not an option"

� Deadlines are sometimes real, sometimes not− become a routine on the US side− easy to pressure over the email− very difficult for a non-empowered team to push back− risk: inflation of urgency

� Better way:− good planning− proper weighing of deadlines and priorities− frequent reporting− local management


Pattern: "Surprises"

� Good news travels better than bad news...− should be the other way around− the "cover up": "let's fix, no need to tell...."− over time: needing bigger cover ups to conceal

smaller ones− not unique for off-shoring, but more difficult to

detect and deal with

� Once a surprise happens:− you will feel frustrated, and betrayed− fix the problems, point out the consequences of

hiding, avoid screaming and flaming

� Better ways:− agree: NO SURPRISES!!− emphasize again and again− train against this− continuously manage, point out− the magic word: transparency

SUPRISES

4/10/2014

80


Pattern: "Ownership"

� Shared responsibility is no responsibility

� Effort-based versus result-based

� On-shore players feel the off-shore team has a result responsibility

� Off-shore team members feel an effort-based responsibility ("work hard")

� Better way:− clear responsibilities and expectations− on-shore ownership for quality control of system under test

• and therefore the tests

− off-shore ownership of producing good tests and good automation− empower according to ownership


Pattern: "Mythical Man Month"

� Fred Brooks classic book, "Mythical man month":− "Assigning more programmers to a project running behind schedule

will make it even later"− "The bearing of a child takes nine months, no matter how many

women are assigned"

� In test automation, there must be clear ownership of:− test design (not just cranking out test cases)− automation, this is different skill and interest

� Assign at least the following roles:− project lead, owns quality and schedule− test lead: owns test design, coaches and coordinates the other testers− automation: make the actions work (assuming ABT, not the test

cases)

� Define distinct career paths in: testing, automation, management

4/10/2014

81


Pattern: "Cooperation"

� Communication is tedious, takes a long time

� Questions, questions, questions, ...− reverse: questions don't get answered

� For at least one side in private time, extra annoying

� Misunderstandings, confusion, actions not followed up− double check apparent "crazy things" with the team before jumping to conclusions, and

actions (assume the other side is not "nuts" or "dumb"...)

� Please understand: distance fosters conflicts− we're born that way, be ready for it

� Better ways:− prioritize training, coaching, preparation and planning. Saves a lot of questions...− write stuff down, use briefs, minutes− define workflows and information flows

• buckets, reporting, select and use good tools

− specialize meetings• table things for in-depth meetings• ask to meet internally first

− be quick, no more than 30 mins


Cooperation

4/10/2014

82


Training, some ideas

� Many areas, big pay-offs:− system under test− subject matter under test, domain knowledge− methods, best practices− technologies, tools, ...− processes− soft skills, like creativity, critical thinking, management, ...− language− cross-cultural

� Have exams− think about the consequences of passing and failing− teams pay more attention when they know they will get tested− you will know whether you were understood

� Also have coaching and train-the-trainers− more experienced people help newbie's− also runs a risk: bad habits can creep in and procreate− "Tribal knowledge", learning by osmosis, water cooler conversations, encourage it− consider "special interest groups (SIG's)"

� Rule of thumb for off-shore teams: hire for technical knowledge, train for business knowledge

� The on-shore staff needs training and coaching too


Additional ideas

� Go there, be with the team, experience yourself how "your side" is doing− I go about twice per year

� Manage ownership, is it you or them− the distinction between efforts and results

� Provide clear direction, constant attention and coaching

� Supervise, supervise, supervise− but don't micromanage if the other side has ownership

� Ask to create example products (like ABT test modules and actions), review them carefully, and use as direction for subsequent work

� Leadership style: participative styles seem most common (as opposed to consensus or authoritative, see also [Al-Ani])

� Organize informal/fun events, provide a good environment− solidify the group, improve retention− include visiting US staff, this tends to do a lot of good ("priceless")

� Manage expectations− stuff takes time and energy− differences can be addressed, but not 100%

cake...

4/10/2014

83


Outsourcing and Agile

� If done well, can provide relieve to a lot of the patterns

� Several models possible

� Model 1: Full team outsourcing− development, testing and automation

� Model 2: "2nd unit"− off-shore team works under control of sprint team members

� Model 3: Part of integrated team:− needs online tool like Jira or Rally− you must have shared meetings− advantage: more project time

� Large scale test development and automation might be easier to outsource than development


Summary

� Not all "big project" challenges are the same

� Think before you do. Best results come from planning well, and combining effective concepts, tricks and tools

� Consider tests and automation as products

� Team work is a key for short term and long term success

� There are many options for infrastructure, but keep an eye on economy and planning

� Off-shoring can help scale up, but needs attention to do it right, in particular communication

4/10/2014

84


Homework . . .

1. Testing Computer Software, Cem Kaner, Hung Nguyen, Jack Falk, Wiley

2. Lessons Learned in Software Testing, Cem Kaner, James Bach, Bret Pettichord, Wiley

3. Experiences of Test Automation, Dorothy Graham, Mark Fewster, Addison Wesley, 2012

4. Automating Software Testing, Dorothy Graham, Mark Fewster, Addison Wesley

5. "Build a Successful Global Training Program", Michael Hackett, www.logigear.com

6. Action Based Testing (overview article), Hans Buwalda, Better Software, March 2011

7. Action Figures (on model-based testing), Hans Buwalda, Better Software, March 2003

8. Integrated Test Design & Automation, Hans Buwalda, Dennis Janssen and Iris Pinkster, Addison Wesley

9. Soap Opera Testing (article), Hans Buwalda, Better Software Magazine, February 2005

10. Testing with Action Words, Abandoning Record and Playback, Hans Buwalda, Eurostar 1996

11. QA All Stars, Building Your Dream Team, Hans Buwalda, Better Software, September 2006

12. The 5% Solutions, Hans Buwalda, Software Test & Performance Magazine, September 2006

13. Happy About Global Software Test Automation, Hung Nguyen, Michael Hackett, e.a., Happy About

14. Testing Applications on the Web, Hung Nguyen, Robert Johnson, Michael Hackett, Wiley

15. Practical Combinatorial Testing, Richard Kuhn, Raghu Kacker, Yu Lei, NIST, October, 2010

16. Agile Software Testing in a Large-Scale Project, David Talby, Arie Keren, Orit Hazzan, Yael Dubinsky, IEEE Software, July/August 2006

17. JMeter in the Cloud, Jörg Kalsbach, http://aws.amazon.com/amis/2924

18. Using Monkey Test Tools, Noel Nyman, STQE issue January/February 2000

19. High Volume Test Automation, Cem Kaner, Walter P. Bond, Pat McGee, StarEast 2004

20. Descriptive Analysis of Fear and Distrust in Early Phases of GSD Projects, Arttu Piri, Tuomas Niinimäki, Casper Lassenius, 2009 Fourth IEEE International Conference on Global Software Engineering [Piri]

21. Quality Indicators on Global Software Development Projects: Does 'Getting to Know You' Really Matter?Olly Gotel, Vidya Kulkarni, Moniphal Say, Christelle Scharff, Thanwadee Sunetnanta, 2009 Fourth IEEE International Conference on Global Software Engineering [Gotel]

The Challenges of BIG Testing: Automation, Virtualization, Outsourcing, and More

Technology

Transcript of The Challenges of BIG Testing: Automation, Virtualization, Outsourcing, and More