ETL Testing

The purpose of Metadata Testing is to verify that the table definitions conform to the data model and application design specifications.

Data Type Check

Verify that the table and column data type definitions are as per the data model design specifications.

Example: Data Model column data type is NUMBER but the database column data type is STRING (or VARCHAR).

Data Length Check

Verify that the length of database columns are as per the data model design specifications.

Example: Data Model specification for the ‘first_name’ column is of length 100 but the corresponding database table column is only 80 characters long.

Index / Constraint Check

Verify that proper constraints and indexes are defined on the database tables as per the design specifications.

1. Verify that the columns that cannot be null have the ‘NOT NULL’ constraint.
2. Verify that the unique key and foreign key columns are indexed as per the requirement.
3. Verify that the table was named according to the table naming convention.

Example 1: A column was defined as ‘NOT NULL’ but it can be optional as per the design.
Example 2: Foreign key constraints were not defined on the database table resulting in orphan records in the child table.

Metadata Naming Standards Check

Verify that the names of the database metadata such as tables, columns, indexes are as per the naming standards.

Example: The naming standard for Fact tables is to end with an ‘_F’ but some of the fact tables names end with ‘_FACT’.

Metadata Check Across Environments

Compare table and column metadata across environments to ensure that changes have been migrated appropriately.

Example: A new column added to the SALES fact table was not migrated from the Development to the Test environment resulting in ETL failures.

Automate metadata testing with ETL Validator

ETL Validator comes with Metadata Compare Wizard for automatically capturing and comparing Table Metadata.

1. Track changes to Table metadata over a period of time. This helps ensure that the QA and development teams are aware of the changes to table metadata in both Source and Target systems.
2. Compare table metadata across environments to ensure that metadata changes have been migrated properly to the test and production environments.
3. Compare column data types between source and target environments.
4. Validate Reference data between spreadsheet and database or across environments.

The purpose of Data Completeness tests are to verify that all the expected data is loaded in target from the source. Some of the tests that can be run are : Compare and Validate counts, aggregates (min, max, sum, avg) and actual data between the source and target.

Record Count Validation

Compare count of records of the primary source table and target table. Check for any rejected records.

Example: A simple count of records comparison between the source and target tables.

Source Query

SELECT count(1) src_count FROM customer

Target Query

SELECT count(1) tgt_count FROM customer_dim

Column Data Profile Validation

Column or attribute level data profiling is an effective tool to compare source and target data without actually comparing the entire data. It is similar to comparing the checksum of your source and target data. These tests are essential when testing large amounts of data.

Some of the common data profile comparisons that can be done between the source and target are:

1. Compare unique values in a column between the source and target.
2. Compare max, min, avg, max length, min length values for columns depending of the data type.
3. Compare null values in a column between the source and target.
4. For important columns, compare data distribution (frequency) in a column between the source and target.

Example 1: Compare column counts with values (non null values) between source and target for each column based on the mapping.

Source Query
SELECT count(row_id), count(fst_name), count(lst_name), avg(revenue) FROM customer

Target Query
SELECT count(row_id), count(first_name), count(last_name), avg(revenue) FROM customer_dim

Example 2: Compare the number of customers by country between the source and target.

Source Query

SELECT country, count(*) FROM customer GROUP BY country

Target Query
SELECT country_cd, count(*) FROM customer_dim GROUP BY country_cd

Compare Entire Source and Target Data

Compare data (values) between the flat file and target data effectively validating 100% of the data. In regulated industries such as finance and pharmaceutical, 100% data validation might be a compliance requirement. It is also a key requirement for data migration projects. However, performing 100% data validation is a challenge when large volumes of data is involved. This is where ETL testing tools such as ETL Validator can be used because they have an inbuilt ELV engine (Extract, Load, Validate) capabile of comparing large values of data.

Example: Write a source query that matches the data in the target table after transformation.

Source Query
SELECT cust_id, fst_name, lst_name, fst_name||’,’||lst_name, DOB FROM Customer

Target Query
SELECT integration_id, first_name, Last_name, full_name, date_of_birth FROM Customer_dim

Automate Data Completeness Testing using ETL Validator

ETL Validator comes with Data Profile Test Case, Component Test Case and Query Compare Test Case for automating the comparison of source and target data.

1. Data Profile Test Case: Automatically computes profile of the source and target query results – count, count distinct, nulls, avg, max, min, maxlength and minlength.
2. Component Test Case: Provides a visual test case builder that can be used to compare multiple sources and target.
3. Query Compare Test Case: Simplifies the comparison of results from source and target queries.

The purpose of Data Quality tests is to verify the accuracy of the data. Data profiling is used to identify data quality issues and the ETL is designed to fix or handle these issue. However, source data keeps changing and new data quality issues may be discovered even after the ETL is being used in production. Automating the data quality checks in the source and target system is an important aspect of ETL execution and testing.

Duplicate Data Checks

Look for duplicate rows with same unique key column or a unique combination of columns as per business requirement.

Example: Business requirement says that a combination of First Name, Last Name, Middle Name and Data of Birth should be unique.
Sample query to identify duplicates 
SELECT fst_name, lst_name, mid_name, date_of_birth, count(1) FROM Customer GROUP BY fst_name, lst_name, mid_name HAVING count(1)>1

Data Validation Rules

Many database fields can contain a range of values that cannot be enumerated. However, there are reasonable constraints or rules that can be applied to detect situations where the data is clearly wrong. Instances of fields containing values violating the validation rules defined represent a quality gap that can impact ETL processing.

Example: Date of birth (DOB). This is defined as the DATE datatype and can assume any valid date. However, a DOB in the future, or more than 100 years in the past are probably invalid. Also, the date of birth of the child is should not be greater than that of their parents.

Data Integrity Checks

This measurement addresses “keyed” relationships of entities within a domain. The goal of these checks is to identify orphan records in the child entity with a foreign key to the parent entity.

1. Count of records with null foreign key values in the child table.
2. Count of invalid foreign key values in the child table that do not have a corresponding primary key in the parent table.

Example: In a data warehouse scenario, fact tables have foreign keys to the dimension tables. If an ETL process does a full refresh of the dimension tables while the fact table is not refreshed, the surrogate foreign keys in the fact table are not valid anymore. “Late arriving dimensions” is another scenario where a foreign key relationship mismatch might occur because the fact record gets loaded ahead of the dimension record.

1. Count of null or unspecified dimension keys in a Fact table: SELECT count(cust_id) FROM sales where cust_id is null
2. Count of invalid foreign key values in the Fact table: SELECT cust_id FROM sales minus SELECT s.cust_id FROM sales s, customers c where s.cust_id=c.cust_id

Automate data quality testing using ETL Validator

ETL Validator comes with Data Rules Test Plan and Foreign Key Test Plan for automating the data quality testing.

1. Data Rules Test Plan: Define data rules and execute them on a periodic basis to check for data that violates them.
2. Foreign Key Test Plan: Define data joins and identify data integrity issues without writing any SQL queries.

Transformation testing using White Box approach

1. Review the source to target mapping design document to understand the transformation design.
2. Apply transformations on the data using SQL or a procedural language such as PLSQL to reflect the ETL transformation logic.
3. Compare the results of the transformed test data with the data in the target table.

Example: In a financial company, the interest earned on the savings account is dependent the daily balance in the account for the month.

1. Review the requirement and design for calculating the interest.
2. Implement the logic using your favourite programming language.
3. Compare your output with data in the target table.

Transformation testing using Black Box approach

1. Review the requirements document to understand the transformation requirements.
2. Prepare test data in the source systems to reflect different transformation scenarios.
3. Come with the transformed data values or the expected values for the test data from the previous step.
4. Compare the results of the transformed test data in the target table with the expected values.

Example: In a financial company, the interest earned on the savings account is dependent the daily balance in the account for the month.

1. Review the requirement for calculating the interest.
2. Setup test data for various scenarios of daily account balance in the source system.
3. Compare the transformed data in the target table with the expected values for the test data.

1. Visual Test Case Builder: Component test case has a visual test case builder that makes it easy to rebuild the transformation logic for testing purposes.
2. Workschema: ETL Validator’s workschema stores the test data from source and target queries. This makes it easy for the tester to implement transformations and compare using a Script Component.
3. Benchmark Capability: Makes it easy baseline the target table (expected data) and compare the latest data with the baselined data.

Example 1: The length of a comments column in the source database was increased but the ETL development team was not notified. Data started getting truncated in production data warehouse for the comments column after this change was deployed in the source system.

Example 2: One of the index in the data warehouse was dropped accidentally which resulted in performance issues in reports.

Automated ETL Testing

Regression testing by baselining target data

1. Execute the ETL before the change and make a copy of the target table.
2. Execute the modified ETL that needs to be regression tested.
3. Compare data in the target table with the data in the baselined table to identify differences.
4. Compare the results of the transformed test data in the target table with the expected values.

Example: In the data warehouse scenario, ETL changes are pushed on a periodic basis (eg. monthly). The tester is tasked with regression testing the ETL. By following the steps outlined above, the tester can regression test key ETLs.

Verify that data conforms to reference data standards

Example: Values in the country_code column should have a valid country code from a Country Code domain.
select distinct country_code from address minus select country_code from country

Compare domain values across environments

Example: Compare Country Codes between development, test and production environments.

Track reference data changes

Example: A new country code has been added and an existing country code has been marked as deleted in the development environment without the approval or notification to the data steward.

Automate reference data testing using ETL Validator

1. Baseline reference data and compare with the latest copy to track changes to reference data.
2. Define data rules to verify that the data conform to the domain values.

ETL process is generally designed to be run in a Full mode or Incremental mode. When running in Full mode, the ETL process truncates the target tables and reloads all (or most) of the data from the source systems. Incremental ETL only loads the data that changed in the source system using some kind of change capture mechanism to identify changes. Incremental ETL is essential to reducing the ETL run times and it is often used method for updating data on a regular basis. The purpose of Incremental ETL testing is to verify that updates on the sources are getting loaded into the target system properly.

While most of the data completeness and data transformation tests are relevant for incremental ETL testing, there are a few additional tests that are relevant. To start with, setup of test data for updates and inserts is a key for testing Incremental ETL.

Duplicate Data Checks

When a source record is updated, the incremental ETL should be able to lookup for the existing record in the target table and update it. If not this can result in duplicates in the target table.

Example: Business requirement says that a combination of First Name, Last Name, Middle Name and Data of Birth should be unique.

Sample query to identify duplicates 

SELECT fst_name, lst_name, mid_name, date_of_birth, count(1) FROM Customer GROUP BY fst_name, lst_name, mid_name HAVING count(1)>1.

Compare Data Values

Verify that the changed data values in the source are reflecting correctly in the target data. Typically, the records updated by an ETL process are stamped by a run ID or a date of the ETL run. This date can be used to identify the newly updated or inserted records in the target system. Alternatively, all the records that got updated in the last few days in the source and target can be compared based on the incremental ETL run frequency.

Example: Write a source query that matches the data in the target table after transformation.
Source Query

SELECT fst_name||’,’||lst_name FROM Customer where updated_dt>sysdate-7
Target Query
SELECT full_name FROM Customer_dim where updated_dt>sysdate-7

Data Denormalization Checks

Denormalization of data is quite common in a data warehouse environment. Source data is denormalized in the ETL so that the report performance can be improved. However, the denormalized values can get stale if the ETL process is not designed to update them based on changes in the source data.

Example: The Customer dimension in the data warehouse is denormalized to have the latest customer address data. However, the incremental ETL for the Customer Dim was not designed to update the latest address data when the customer updates their address because it was only designed to handle the Change Capture on the Customer source table and not the Customer_address table. The Customer address shown in the Customer Dim was good when a Full ETL was run but as the Customer Address changes come in during the Incremental ETL, the data in the Customer Dim became stale.

Source Query
SELECT cust_id, address1, address2, city, state, country FROM Customer SELECT cust_id, address1, address2, city, state, country,
ROW_NUMBER( ) OVER (PARTITION BY cust_id ORDER BY created_date NULLS LAST) addr_rank
FROM Customer
WHERE ROW_NUMBER( ) OVER (PARTITION BY cust_id ORDER BY created_date NULLS LAST) = 1

Target Query
SELECT cust_id, address1, address2, city, state, country FROM Customer_dim

Slowly Changing Dimension Checks

While there are different types of slowly changing dimensions (SCD), testing of and SCD Type 2 dimension presently a unique challenge since there can be multiple records with the same natural key. Type 2 SCD is designed to create a new record whenever there is a change to a set of columns. The latest record is tagged with a flag and there are start date and end date columns to indicate the period of relevance for the record. Some of the tests specific to a Type 2 SCD are listed below:

1. Is a new record created every time there is a change to the SCD key columns as expected?
2. Is a latest record tagged as the latest record by a flag?
3. Are the old records end dated appropriately?

Automate incremental ETL testing using ETL Validator

ETL Validator comes with Benchmarking Capability in Component Test Case for automating the incremental ETL testing. Benchmarking capability allows the user to automatically compare the latest data in the target table with a previous copy to identify the differences. These differences can then be compared with the source data changes for validation.

End-to-End Data Testing

1. Setup test data in the source system.
2. Execute ETL process to load the test data into the target.
3. View or process the data in the target system.
4. Validate the data and application functionality that uses the data.

Example: Let’s consider a data warehouse scenario for Case Management analytics using OBIEE as the BI tool. An executive report shows the number of Cases by Case type in OBIEE. However, during testing when the number of cases were compared between the source, target (data warehouse) and OBIEE report, it was found that each of them showed different values. As part of this testing it is important to identify the key measures or data values that can be compared across the source, target and consuming application.

Automate integrated ETL testing using ETL Validator

Example 1: A lookup might perform well when the data is small but might become a bottle neck that slowed down the ETL task when there is large volume of data. What can make it worse is that the ETL task may be running by itself for hours causing the entire ETL process to run much longer than the expected SLA.

Example 2: An incremental ETL task was updating more records than it should. When the data volumes were low in the target table, it performed well but when the data volumes increased, the updated slowed down the incremental ETL tremendously.

End-to-End Data Testing

1. Estimate expected data volumes in each of the source table for the ETL for the next 1-3 years.
2. Setup test data for performance testing either by generating sample data or making a copy of the production (scrubbed) data.
3. Execute Full ETL process to load the test data into the target.
4. Review each individual ETL task (workflow) run times and the order of execution of the ETL. Revisit ETL task dependencies and reorder the ETL tasks so that the tasks run in parallel as much as possible.
5. Setup test data for incremental ETL process with the data change volumes as expected during an incremental ETL.
6. Executing incremental ETL. Review ETL task load times and the order of execution of the tasks to identify bottlenecks.