Upload
porter-lobdell
View
220
Download
4
Tags:
Embed Size (px)
Citation preview
High Volume Test Automation in PracticeAndy TinkhamPrincipal Lead Consultant, QATMagenic Technologies
Acknowledgements» This presentation draws on the knowledge shared by the attendees of
WTST 12 in Melbourne, FL (Jan 25-27, 2013, hosted by the Harris Institute for Assured Information at the Florida Institute of Technology and Kaner, Fiedler & Associates, LLC)» Cem Kaner, Catherine Karena, Michael Kelly, Rebecca Fiedler, Janaka Balasooriyi,
Thomas Bedran, Jared Demott, Keith Gallagher, Doug Hoffman, Dan Hoffman, Harry Robinson, Rob Sabourin, Andy Tinkham, Thomas Vaniotis, Tao Xie, Casey Doran, Mark Fiorvanti, Michal Frystacky, Scott Fuller, Nawwar Kabbani, Carol Oliver, Vadym Tereschenko
» This material is heavily drawn from Cem Kaner’s blog posts on kaner.com and context-driven-testing.com, referenced at the end of this slide deck
About me» 17 years in testing industry» Principal Lead Consultant at Magenic Technologies» Doctoral student at Florida Tech
» Host free virtual office hours roughly weeklyhttp://ohours.org/andytinkham
» http://magenic.com/Blog.aspx» http://testerthoughts.com» http://twitter.com/andytinkham
What is High Volume Test Automation (HiVAT)?
“A family of test techniques that enable a tester to run & evaluate
arbitrarily many computer-assisted tests”
-- WTST 12 working definition
Let’s break that down…• Many ways to
do HiVAT• Different ways
for different goals
Family of test techniques
• Not replacing a human
• Augmenting a tester’s skill set
Enable a tester
• Need executable tests
• Need some sort of oracle
Run & evaluate
• Easy to change number of tests
• Not 1:1 matchup with manual tests
Arbitrarily Many
• Continuum of manual & automated
• Different tests at different spots
Computer-assisted tests
Manual & automated testsE
very
test
has
man
ual ele
men
ts • A human designs it• A human wrote the
code• A human analyzes
the results Every
test
has
au
tom
ate
d
ele
men
ts • Transforming the inputs to outputs is done by the computer
Every test falls somewhere on a continuum between the two extremes
HiVAT tests tend toward the automated side» Human still designs overall tests (possibly very high-level)» Computer may determine inputs, paths and expected results» Computer evaluates individual results» Human determines stopping criteria
» Number of tests» Time» First bug
» Human analyzes overall results
…but are different from “traditional” automation
» Include many iterations of execution» May run for longer periods of time» Sometimes involve more randomness» Can be focused on looking for unknown risks rather than identified risks
Why do HiVAT?» Find problems that occur in only a small subset of input values» Find difficult to encounter bugs like race conditions or corrupted state» Catch intermittent failures» Leverage idle hardware
» Address risks and provide value in ways that traditional automation & manual testing don’t normally do
How do we do HiVAT?» Lots of ways!
» Kaner gives this classification scheme which covers many techniques (including the ones we’re about to talk about)
Focus on Inputs
Exploit available
oracle
Exploit existing tests or
tools
Methods that focus on inputs» Testers usually divide inputs into equivalence classes and pick high-value
representative values» For reasonably-sized datasets, automation doesn’t need to do this!» Run all (or at least many of) the values through the automation
» Alternatively, use random input generation to geta stream of input values to use for testing
Parametric Variation» Replace small equivalence class representative sets» Some input sets may allow running the total set of inputs
» Doug Hoffman’s MASPAR example» Others may still require sampling
» Valid passwords example» Sampling can be optimized if data is well understood» Can generate random values
High-Volume Combination Testing» Testers often use combinatorial test techniques to get a workable set of
combinations to cover interactions» These techniques leave combinations uncovered
» If we know which uncovered combinations are more important or risky, we can add them to the test set
» What about when we don’t know which ones are of interest?» HiVAT tests can run many more combinations through than are usually done» Sampling can be same as Parametric Variation
» Retail POS system example
Input Fuzzing/Hostile Data Stream Testing» Given a known good set of inputs» Make changes to the input and run each changed values through the
system» Watch for buffer overruns, stack corruption, crashes, and other system-
level problems
» Expression Blend example» Alan Jorgensen’s Acrobat Reader work
Automated Security Vulnerability Checking» Scan an application for input fields» For each input field, try a variety of common SQL Injection and Cross-Site
Scripting attacks to detect vulnerabilities
» Mark Fiorvanti’s WTST paper (see references)
One problem with input focused tests» We need an oracle!
» It can be hard to verify the correctness of the results without duplicating the functionality we’re testing
» Input-focused tests may look for more obvious errors» Crashes» Memory problems» Simple calculations
Methods that exploit oracles» Sometimes we already have an oracle available» If so, we can take advantage of it!
Functional Equivalence» Run lots of inputs through the SUT and another system that does the
same thing, then compare outputs
» FIT Testing 2 exam example
Constraint Checks» Look for obviously bad data
» US ZIP codes that aren’t 5 or 9 digits long» End dates that occur before start dates» Pictures that don’t look right
State-Model Walking» 3 things required
» State model of the application» A way to drive the application» A way to determine what state we’re in
Methods that exploit existing tests or tools» Existing artifacts can be used in high-volume testing
» Tests» Load Generators
Long-Sequence Regression Testing» Take a set of individually passing automated regression tests» Run them together in long chains over extended periods of time» Watch for failures
» Actions may leave corrupted state that only later appears» Sequence of actions may be important
» Mentsville example
High-Volume Protocol Testing» Send a string of commands to a protocol handler
» Web service method calls» API calls» Protocols with defined order
Load-enhanced Functional Testing» Run your existing automated functional tests AND your automated load
generation at the same time» Add in additional diagnostic monitoring if available
» Systems behave differently under load» System resource problems may not be
visible when resources are plentiful» Timing issues
Starting HiVAT in your organization» Inventory what you already have
» Existing tests you can chain together (Preferably without intervening clean-up code)
» Tools you can put to additional uses» Oracles you can use» Places where small samples have been chosen from a larger data set» Hardware that is sometimes sitting idle
Starting HiVAT in your organization» Match your inventory up to techniques that can take advantage of them» Think about what sorts of risks and problems a technique could reveal in
your application» For each risk, do you have other tests that can be reasonably expected to
cover that issue?» How much value is there in getting information about the risk?» How much effort is required to get the information?» What other tasks could you do in the same time? » Is the value of the information ≥ the cost to implement + the value of the
other tasks?
Summary» High volume automated testing is a family of test techniques focused on
running an arbitrary number of tests» The number of tests is often defined by an amount of time or coverage of
a set of values rather than trying for a minimal set » Some high-volume techniques focus on covering a set of inputs» Some take advantage of an accessible oracle» Some reuse existing artifacts in new ways» Determining what makes sense for you is a matter of risk and value
References» Cem Kaner’s High Volume Test Automation Overview
http://kaner.com/?p=278
» Cem’s WTST 12 write-uphttp://context-driven-testing.com/?p=69
» WTST 12 home page (with links to papers and slides, including Mark Fiorvanti’s)http://wtst.org
» Doug Hoffman’s MASPAR examplehttp://www.testingeducation.org/BBST/foundations/Hoffman_Exhaust_Options.pdf
» Alan Jorgensen’s “Testing With Hostile Data Streams” paperhttps://www.cs.fit.edu/media/TechnicalReports/cs-2003-03.pdf
» Pat McGee & Cem Kaner’s Long-Sequence Regression Test (Mentsville) planhttp://www.kaner.com/pdfs/MentsvillePM-CK.pdf
Contact Information
Andy Tinkham
Magenic Technologies
http://magenic.com
http://ohours.org/andytinkham
http://testerthoughts.com
http://twitter.com/andytinkham