Data Warehouse ETL Testing Best Practices
Data Warehouse ETL Testing Best Practices
Data Warehouse ETL Testing Best Practices
Abstract
One of the greatest risks to success any company implementing a business intelligence
system can make is rushing a data warehouse into service without testing it effectively.
Even wise IT managers, who follow the Old Russian proverb, "trust, but verify," need to,
maintain their vigilance. There are pitfalls in the testing process, too.
Many organizations create test plans and assume that testing is over when every single test
condition passes their expected results. In reality, this is a very difficult bar to meet. Often,
some requirements turn out to be unattainable when tested against production information.
Some business rules turn out to be false or incredibly more complex than originally thought.
Data warehousing applications keep on changing with changing requirements. This white paper shares
some of the best practices of the experiences of data warehouse testing.
“If you torture data sufficiently, it will confess to almost anything.”
We know how critical the data is in a data warehouse when it integrates data from
different sources. For example, in the healthcare industry, it helps users to answer
business questions about physicians, plan the performance, market share and geographic
variations in clinical practice, health outcomes etc. Thus if the data is so sensitive, critical
and vast, we can understand how much challenging it would be. Thus this is a menial
effort to write about some of the best practices we learned while doing it on ground to
share it with others.
How much confident a company can be to implement its data warehouse in the market
without actually testing it thoroughly. The organizations gain the real confidence once the
data warehouse is verified and validated by the independent group of experts known as
“Data warehouse testers”.
Many organizations don’t follow the right kind of testing methodology and are never sure
about how much testing is enough to implement their data warehouse in the market. In
reality, this is a very difficult bar to meet. Often, some requirements are difficult to test.
Some business rules turn out to be erroneous, wrongly understood or highly complex than
originally thought.
www.pureconferences.com
Pure Conference Solution Pvt Ltd., A-108B, Sector 58, NOIDA 201301, India; Tel: +91 120 4621080
Page 1 of 6
Data warehousing applications keep on changing with changing requirements. This white
paper explains the need to have a data warehouse application testing in place and it
mentions the various steps of the testing process covering the best practices.
Most of the organization runs their businesses on the basis of collection of data for
strategic decision- making. To take a competitive edge the organization should have the
ability to review historical trends and monitor real-time functional data.
Most importantly, we never appreciate if the bug is detected at the later stage of testing
cycles because it could easily lead to very high financial losses to the project. So data
warehouse testing following the best practice is unavoidable to remain at the top of the
business.
1) Business understanding
a. High Level Test Approach
b. Test Estimation
c. Review Business Specification
d. Attend Business Specification and Technical Specification
walkthroughs
When we test, we take sample data from the designed architecture and the test data files
are usually provided to the testers by the developers. Of course, the test data should be
www.pureconferences.com
Pure Conference Solution Pvt Ltd., A-108B, Sector 58, NOIDA 201301, India; Tel: +91 120 4621080
Page 2 of 6
able to cover all the possible scenarios w.r.t the requirements while we do the predictions
to define the expected test case results, however if the data provided to us is not
supportive enough to cover all the business rules then we go for the data mocking that is
explained in the later section of the paper.
Before we move further let me share a very interesting example which we came across
recently when performing our testing.
On May, 2008 our Data warehouse implemented the logic to use a field, Calendar
Year/Plan Year (field name) which was passed by the Source, for which the source
considered the Policy year as 'Y' and Calendar year as 'N', however the ETL used
different notation to populate Calendar Year and Policy Year details on the target table,
as the logic change was not properly informed. After the project was implemented it was
discovered that since there was a miscommunication between source system and
Business Analysts, it did not recognize the change and millions of rows which were
populated with incorrect data, i.e.
The expected data was Policy year = Y and
Calendar year = N
The actual data was Policy year = N and
Calendar year = Y
Because of wrong notation, data is behaving oddly and in reporting, it leads to huge loss
to the project. Thus effective communication as a best practice could have helped prevent
this situation.
4 What is ETL?
ETL stands for extract, transform, and load. It can consolidate the scattered data for any
organization while working with different departments. It can very well handle the data
coming from different departments.
ETL can transform not only data from different departments but also data from different
sources altogether. For example, any organization is running its business on different
environments like SAP and Oracle Apps for their businesses. If the higher management
wants to take discussion on their business, they want to make the data integrated and used
it for their reporting purposes. ETL can take these two source system data and make it
integrated in to single format and load it into the tables.
www.pureconferences.com
Pure Conference Solution Pvt Ltd., A-108B, Sector 58, NOIDA 201301, India; Tel: +91 120 4621080
Page 3 of 6
5 How is data warehouse testing different from normal testing?
The main difference in testing a data warehouse (DW) is that we basically involve the
SQL queries in our test case documents. It is vital to test both the initial loads of the Data
Warehouse from the source i.e. when it gets extracted and then updating it on the target
table i.e. the loading step. In specific cases, where trouble shooting is required, we verify
intermediate steps as well.
A defect or bug detection can be appreciated if and only if it is detected early and is fixed
at the right time without leading to a high cost. So to achieve it, it is very important to set
some basic testing rules. They are:
• No Data losses
• Correct transformation rules
• Data validation
• Regression Testing
• Oneshot/ retrospective testing
• Prospective testing
• View testing
• Sampling
• Post implementation
We are now going to talk with reference to the practices/strategies we implement in our
current project on each of them.
6 No Data losses
• We verify that all expected data gets loaded into the data warehouse. This
includes validating that all records, all fields and the full contents of each field are
loaded without any truncation occurs at any step in the process.
• As and when required negative scenarios are also validated. Some of the
examples are validating special characters etc
www.pureconferences.com
Pure Conference Solution Pvt Ltd., A-108B, Sector 58, NOIDA 201301, India; Tel: +91 120 4621080
Page 4 of 6
7 Correct transformation Rules
• We ensure that all data is transformed correctly according to business rules, it
could be straight move, simple transformation or the complex transformation.
• The best method could be to pick some sample records, use the “EXACT”
formula in the excel worksheet and compare the results to validate data
transformations manually. This should be ideally done once the testers are
absolutely clear about the transformation rules in the business specification.
8 Data Validation
• We say we have achieved the quality when we successfully fulfill customer’s
requirements. In other words we basically achieve a value for our customer. Since
in data warehouse testing; the test execution revolves around the data, so it is
important to achieve the degree of excellence for the data and for that we do the
data validation for both the data extracted from the source and then getting loaded
at the table.
4 Citations
The surnames of authors and year of publication should be given to the corresponding
text. The references should be left aligned in 6-point spacing (after) and 0-point spacing
(before) should be there for each reference.
• Journal articles:
Cooper, V., Anderson, A.: 1997: The Conference Proceedings. WIT Magazine; 1,
108-121.
• Book
Cooper, Victor: 2008. Noun & Verb Technique. Delhi: Pure Publications.
• Book Chapter
www.pureconferences.com
Pure Conference Solution Pvt Ltd., A-108B, Sector 58, NOIDA 201301, India; Tel: +91 120 4621080
Page 5 of 6
Cooper, Victor: 2009. Understanding exploratory method of software testing. In
Cooper V, Borah J (eds) The Paradigms of Testing: a guide to effective testing
methodologies (pp 27-55). Delhi: Pure Publications.
References
[1] Cooper, V., Anderson, A.: 1997: The Conference Proceedings. ASP.net Magazine;
1, 108-121.
[2] Cooper, Victor: 2008. Noun & Verb Technique. New York: Pure Publications.
Test2008: 2008: Conference Proceeding Template.
http://test2008.in/icsd-template.doc
www.pureconferences.com
Pure Conference Solution Pvt Ltd., A-108B, Sector 58, NOIDA 201301, India; Tel: +91 120 4621080
Page 6 of 6