Data Warehouse and Business Intelligence Testing:

Challenges, Best Practices & the Solution

Prepared by
datagaps
http://www.datagaps.com
http://www.youtube.com/datagaps
http://www.twitter.com/datagaps

Contact
[email protected]

Contributors
Narendar Yalamanchilli
Adi Buddhavarapu
Table of Contents
Executive Summary..................................................................................................................................... 1

The problem ................................................................................................................................................. 2

Issues with Data Entities ....................................................................................................................................................... 3
Issues with Data Warehouses ............................................................................................................................................. 4
Issues with Business Intelligence Applications ............................................................................................................ 5

Best practices ............................................................................................................................................... 6

Data Entity Testing.................................................................................................................................................................. 6
Data Warehouse Testing ...................................................................................................................................................... 6
Business Intelligence Testing .............................................................................................................................................. 8

Testing Metrics .......................................................................................................................................... 11

The solution ............................................................................................................................................... 13

EXECUTIVE SUMMARY
Testing data warehouse and business intelligence projects is challenging due to the
heterogeneous nature of data sources and large volumes of data and reports. Based
on our experience in implementing various data warehouse and business
intelligence projects, we have come up with a set of best practices/patterns in
testing to minimize the risks associated with data intensive projects. Our
methodology conceptually divides testing into three areas.
Data Entity Testing: This is to ensure that the data in every source (flat files,
tables etc) or target entities (fact, dimension etc) is as expected.
ETL Testing: This is to ensure that the data moved from the source systems
to the target/warehouse as expected.
BI Testing: This is to ensure that teams have a firm grasp on reports and
dashboards from a regression, performance and stress testing stand point.
This paper walks the reader through the typical issues which are encountered in
testing warehouses and suggests best practices for automation

Page 1
THE PROBLEM
Data warehouse projects often involve extracting large volumes of data from
multiple data sources before transforming and loading. Testing of data
warehouse requires a data-centric testing approach which is different from
testing of traditional transactional applications. Most of the current test
automation tools are focussed on transaction application testing and are not
designed for testing of data warehouses. Hence it is a common practice for
Business Analysts, Engineers and QA Teams to manually test the ETL by
copying a handful of records from source and target tables into an excel spread
sheet and compare the data. This is a tedious and error prone process when
done manually for large number of ETL. Testing is further complicated because
the ETL process can run in an incremental mode or full mode often making it
necessary to create separate sets of test cases for these two modes. Every
incremental change to the source or target system increases the scope for
regression and these manual tests have to be repeated every time.

When we extend this problem to BI environments, testing becomes even more

challenging as there are hundreds of dashboards and reports to be verified.
Testing is further complicated due to the abstraction provided by the BI tool
metadata model because the scenarios for adhoc report requests are not fixed
and testing all the scenarios is not practical. As the business intelligence
application matures, it becomes very difficult to assess the impact of the changes
in the metadata model on existing reports because of the effort involved in
testing all the reports. Most of the current forms of testing involve manual spot
checking by running a few reports and comparing it to the data queried from the
physical data sources (e.g. databases). Often there is very little or no regression
testing performed for any changes being made for these applications since there
is no easy way to automate this testing. The impact of not regression testing
reports can vary from throwing errors, showing wrong data or bad performance
which is usually detected only when the changes go to production. In addition,
system upgrades such as upgrades to BI tool, database or hardware may result in
report accessibility issues, errors in rendering reports or performance issues.

These issues often result in increased cost of ownership, inaccurate insights due
to bad data/reports and the loss of trust in enterprise data and reporting
capabilities. The below section identifies the kinds of issues we see based on the
above categories and the best practices to address them.

We believe that there are three areas to ensure proper data testing.
First, testing has to be done at each data entity level to ensure that the
data coming into the warehouse is as expected.
Second, as data moves from various sources to the warehouse, we need
to ensure that data is exactly as expected.
Finally, the reporting system itself has to be tested for accuracy,
regression, performance and scalability.

Page 2
Issues with Data Entities
Data can enter an enterprise from various sources such as transactional systems,
XML files, flat files, web services etc and it is possible for data to take up
unexpected forms in this step itself. Few examples are below:
1. An application does not have proper data validation and thus allows inaccurate
data to be entered into the system:
a. Invalid email addresses without an @ or not following the email
address pattern
b. A column to store country names has values that do not fall in the
encoded list of values defined for it
c. SSN is a required data attribute but has null values or less than 9
numbers
d. Orphan order item (child) records are persisted even though the
corresponding order (parent) record has been deleted
e. Duplicate records are created in the system for the same customer
f. Data for start and end dates is not consistent because end date is less
than start date
g. Address entered by the customer is a not a valid one or postal code is
null for US addresses

2. A number of advertising companies get customer data on a daily basis in flat

files from various stakeholders. It is possible that the column with the First
Name of the first one million incoming customer records is missing. If a
campaign is sent out using this data set to all the one million customers, they
get emails without their first names in the campaign and may negatively impact
the brand value.
a. Flat file does not contain expected number of data columns
b. Data in a column has invalid type or length eg. Invalid date format or
text in a number column
c. Date of birth is more than 150 years old
d. Address fields are null when country is US when they are expected to
be available
e. The product codes passed in the flat file does not exist in the list stored
in the product code table

Prior to moving this data into other systems within the enterprise, it is important
that we identify such discrepancies early on. The current Data Quality tools
primarily focus on data discovery using data profiling and cleaning of data but do
not provide a good means to measure the quality of the data by providing
capability to define data rules. These data quality tools are often very expensive and
difficult to implement.

Page 3
Issues with Data Warehouses
Data warehouses (and Data Integration) projects typically involve movement of
large volumes of data from multiple data sources.

A small sample of issues noticed in data warehouses are listed below:

1. Sales fact does not have all the sales records from the source because of a
bad join in the query for extracting source data extract
2. Customer Last Name is truncated because the column data lengths are not
matching between the source and target
3. One of the columns in Product dimension is empty due to a defect in ETL
4. Revenue fact has duplicate records due to which the revenue numbers in
the report are off
5. Issues with data transformations result in unexpected data. For example:
a. Lookup transformation for getting country name based on country
code results in null values
b. Foreign keys for Customer dimension are not mapping as expected
in the Sales fact
c. Order amount in the aggregate table does not match the amounts
from the Order fact
d. Customer Status value derived in the ETL is shown as Inactive for
a specific record when the expected value is Active
6. Incremental ETL is not updating the values in the de-normalized address
columns of the Customer dimension with the latest address when the
address changes
7. Incremental ETL did not process changes for the Case data in the source
because the issues with the change capture logic or defect in ETL update
strategy
8. Incremental ETL did not create a new record for a Type-2 slowly changing
dimension
9. Historical Sales data that is expected to be unchanged is different because
of an ETL defect
10. Data is missing in the target table because they are wrongly tagged as error
records
11. Duplicate records are loaded into Customer dimension because of two
different source (ERP and CRM) system resulting in bad data in the reports
Page 4
Issues with Business Intelligence Applications
Business Intelligence applications present a different set of challenges in testing
because they empower the end user to create their own reports and dashboards on
the fly.

Some of the issues noticed during in business intelligence applications are:

1. Defect in ETL for Sales fact results in wrong sales numbers to be shown in
Sales Dashboard
2. Adding a new conformed Product Category dimension in the BI tool
metadata results in extra unwanted joins for reports based out of Sales
subject area thus skewing the sales numbers
3. Business Analyst does not trust the Revenue numbers shown in the revenue
dashboard because they do not match numbers from the source
4. Incomplete changes to Sales subject area are migrated to the next
environment as part of a fix made for Revenue subject area because of
common metadata model (RPD)
5. Developer modifies the measure in Order Trend report without notifying
the business analyst or quality analyst which goes untested to production
6. BI tool upgrade results in bad report performance and errors in some of the
dashboard pages
7. Database upgrade results in performance issues because the queries are
executing with a new plan
8. Rollout of the new Customer Intelligence application to 400 new users
results in all dashboards to run slow because of a network bottleneck
9. A developer modifies column formatting for date column for all subject
areas without realizing the impact
10. A developer renames Employee Name to Owner Name without realizing
that there are existing reports using this attribute
11. ETL developer increases the length of lst_name column or renames it to
last_name without notifying the report developer
12. A change in the initialization block for authorization results in unwanted
loss of access to Customer Management dashboard to business analysts
13. Aggregate numbers shown in the summary report do not match the
numbers shown in the drilldown report

Page 5
 BEST PRACTICES
Testing data warehouses and business intelligence applications requires a data-
centric testing approach. This section provides a list of the type of tests and the
strategy/best practice for doing the test

Data Entity Testing

The focus of data entity testing is to identify the issues with data that is not
following the data rules defined for it.

Type Description Strategy

Data Validity Are the values within the acceptable For each attribute with a defined list of
subset of an encoded list? values, check for values in the data entity
to identify the values that are not in the list
of valid values
Data Are there any unexpected duplicates? Write queries to check for duplicates based
Consistency on combinations of columns that are
expected to be unique
Referential Are there any orphan records or Identify parent-child and many-to-many
Integrity missing foreign keys relationships between data entities and
write queries to identify inconsistencies
Data Validity Is the data as per business rules? For each attribute, run SQL queries to
check the data. This may involve using
Email column does not have valid
regular expressions.
email addresses
Birth date has values older than 200
years
Postal Code has non-numeric values
for US addresses
SSN has null values for some
records
Combination of Customer Name,
address columns is unique

Data Warehouse Testing

The goal of data warehouse (or data integration) testing is to verify that data moved
from source to the target and the transformation rules have been applied as expected.

Type Description Strategy

Data Are the counts matching between While it is ideal to compare the entire
Completeness source and the target? source and target data, it is often a very
time consuming and resource intensive
For example, source records can be
process. Hence comparing the counts is a
missing in target due to bad filter,
good strategy. This will also help identify
rejected records, wrong data types.

Page 6
 duplicate introduced during the ETL.

Compare source and target table row

counts. Compare count of rows with data
in individual columns between source and
target. Compare distinct list of values and
count of rows for each value between the
source and target.
Data Integrity Is the target data consistent with the Data completeness does verify that the data
source data? counts are matching but it does not ensure
that the data values are consistent with the
source. For example, data can get truncated
in a column. Hence comparing the data
values between the source and target for a
sample of data (eg. 10% of the data) can
identify these issues. For testing
incremental runs, the data that got
modified in a specific time period can be
verified. This type of test is appropriate for
data with little (eg. Expression
transformation) or no transformations.
Data Integrity Is the Fact to dimension table For each fact dimension foreign key
foreign key mapped in the ETL Check the count of mapped rows in source
appropriately? DB and compare it with target DB.
For a sample set of fact records, compare
the foreign key mapping between the
source and target DB
Data How to test Lookup Write SQL query to compare the list of
Transformations transformation? values in a column with the list of valid
values expected
Data How to test Aggregate Write SQL query to aggregate the data and
Transformations transformation? compare it with data in aggregate target
table

Data How to test Expression Write SQL query with the expected
Transformations transformation? For eg, are nulls transformation on the source data and
replaced appropriately? compare it with data in target table
Data My transformations are too complex Setup test data in the source for different
Transformations to test using single source and target conditions. Verify that the data in the target
queries against expected result
Data How to perform regression testing Benchmark the target table records for data
Transformations of transformations? that is not expected to change (eg. 10000
records) and compare it after the ETL
change

Page 7
 Business Intelligence Testing
While data warehouse (ETL) testing is an essential part of ensuring that the data
shown in the reports is correct, issues in the BI tool metadata model can cause this
data to be reported wrongly. Hence the primary focus of BI testing is to ensure that
numbers and data shown in reports are correct when compared to the source and
the reports are accessible to the end user from a performance and security
standpoint. Listed below are some of the common test scenarios and strategies.

Type Description Strategy

Unit Testing Dashboard prompts are showing For each prompt in a dashboard page,
wrong data or they are not getting check if the list of values shown is correct
applied to reports appropriately where applicable. Apply all the prompts and
verify the report and the corresponding
physical SQL. Verify that all prompts are
getting applied appropriately to the
drilldowns.
Unit Testing Dashboard page does not conform to Verify that the report layout, prompts,
UI Standards titles, download and filter display meet the
design
Unit Testing The SQL Query generated by BI Write your own SQL query on the target
Server for a report is complex and database and compare the output with the
shows wrong data (numbers) report data
Unit Testing An RPD change for a new subject Compare the SQL Query generated by the
area changes the join conditions for report before and after making RPD
an existing report changes to the RPD
Unit Testing Mismatches between Summary and Aggregate the detail reports output and
detail report data compare it with the summary report data.
Check the links to detail report from charts,
data and table headings in the summary
report
Unit Testing A defect in the new ETL results in Benchmark the report output and compare
the corresponding report showing it. Another option to compare the data
wrong data shown in reports with the output of a query
against the source database
Unit Testing Webcat merge from local desktop to Compare the report data between the local
development environment is and development environments
unsuccessful
Unit Testing How can we test adhoc reporting 1) For each dimension folder, pick all
since the number of possible test attributes and run reports. The objective is
cases can be unlimited? to check if any attribute is not mapped
properly. Also check the dimension
While it is not practical to test all hierarchy.
combinations of adhoc reports is 2) Run report by picking one attribute from
there a method to testing adhoc each dimension and one measure to verify
reports? the joins. Watch for any unwanted joins. If

Page 8
 you are seeing a complex query it is usually
because of bad model or design.
3) Verify that the list of measures and their
names are meaningful
Unit Testing RPD merge results in duplicate Compare the list of presentation layer
attributes attributes before and after the merge
Regression A developer knowingly or Take screenshots of the dashboard pages
Testing unknowingly makes a dashboard UI before and after the changes and compare
change that the business them
analyst/tester is not aware of
Regression Some of the dashboard pages Benchmark and compare the dashboard
Testing suddenly start running slow after a page run time and corresponding SQL
recent release or system change query execution plan before after the
change
Regression An ETL change results in the Benchmark and compare the report data
Testing measure values in the reports to be and chart before after the change
wrong after the full ETL run
Regression Name change of an attribute in the Search for impacted reports in production
Testing presentation layer adversely impacts usage tracking data or search in the web
user reports in production catalog
Regression It is difficult to prioritize which Regression test most commonly used
Testing reports or adhoc request to regression reports or adhoc requests based on
test production usage tracking data
Regression New reports or dashboard pages are Capture dashboard page run times after
Testing running really slow and not meeting disabling caching. Alternatively capture the
SLAs average run times by running them multiple
times at different times
Regression Dashboard page runs fine for some Capture dashboard page run times after
Testing roles but is really slow for other roles disabling caching for users with different
(eg. CEO) roles
Regression Drilldown on Hierarchy columns is Verify the drilldown time for each hierarchy
Testing very slow column with caching disabled
Multi- A recent webcat migration results in Compare the report access for key users
Environment access issue due to which some of the with different roles across environments
users cannot access their reports after the migration
Multi- Uptake of new BI releases (eg. 10g to Have a pre and post upgrade environments
Environment 11g or 11.1.1.5 to 11.1.1.7) results in pointing to the same database and compare
report data, security and performance the report data and performance between
issues the two environments
Security Users see data in reports that they are Test data security by comparing the data for
not supposed to have access the same report between two different users
with different roles
Security Users lose access to reports and Verify the access to reports and dashboards

Page 9
 dashboards that they previously had for users with different roles before and
access to after the security changes or webcat
migration
Stress Dashboard pages start running slow Stress test by simulating concurrent users
after rollout to new users or addition and compare dashboard page performance
of new data to see if the SLA requirements are met
Stress A software or hardware upgrade Stress test by simulating concurrent users
adversely impacts the dashboard page and compare dashboard page performance
performance to see if the SLA requirements are met
Stress Reports run fine in test environments Simulate realistic user load in a test
but run slow in production environment based on production usage
tracking data

Page 10
 TESTING METRICS

Data Entity Test Coverage

Table No. of Rules % of attributes with rules No of attributes

User 0 0 12

Customer 5 10% 25

Sales 12 75% 15

Key Relationship tests

Key Relationship Number of Orphans % of Orphans

Sales-Customer 0 0

Sales-Product 10 1%

Data Rules

Table Column Rule Number of non- % of non-

conforming values conforming values

Customer Postal Postal Code is numeric 4 0.04%

Code where country code = US

Customer Customer Customer Name, Address is 0 0%

Name unique

Customer SSN SSN is not null 0 0%

Customer Email Email like REGEXP ^[a-zA- 1000 10%

Z0-9._%+-]+@[a-zA-Z0-9.-
]+\.[a-zA-Z]{2,4}$

Customer Gender Gender Code in valid List of 100 1%

Code values

Target Table ETL Test Coverage

Page 11
Target Table Data Data Data Data Total
Integrity Completeness Consistency Transformation
Tests Tests Tests Tests

Costs 1 1 0 0 2

SalesRep 0 0 0 0 0

Customer 0 1 1 0 2

Sales 1 0 0 0 1

Dashboard Test Coverage

Dashboard Page Usage Rank No. of Regression Tests

Sales Sales Trends 2 1

Sales Product Profitability 5 1

Sales Sales Details 10 0

Report Test Coverage

Dashboard Page Report Usage Rank No. of Unit Tests

Sales Sales Trends Sales Trends 2 1

Sales Product Profitability Product Profitability 5 1

Sales Sales Details Sales Details 10 0

Page 12
THE SOLUTION
Datagaps offers two solutions that collectively enable comprehensive data warehouse
and business intelligence testing automation. Testing teams, Business Analysts and
Engineers can leverage our solutions to minimize testing costs, expedite time to
market and deliver high quality dashboards and reports to business users. Details of
the solutions are below:
ETL Validator: This primarily addresses Data Entity Testing and ETL Testing. Key
features include:
Wizard driven test case creation
Enterprise grade test case visibility
Enterprise collaboration
Flat file testing automation
Data entity testing automation
BI Validator: This addresses BI Testing Automation. Key features include:
Dashboard testing
Report testing
Multi Environment testing
Upgrade testing
Stress testing

Data sheets for both the products are available on our website for your reference.

Page 13