ETL stands for Extract, Transform, and Load. It is an important concept in Data Warehousing systems. Extraction stands for extracting data from different data sources such as transactional systems or applications. Transformation stands for applying the conversion rules on data so that it becomes suitable for analytical reporting. The loading process involves moving the data into the target system, normally a data warehouse.
The three layers involved in an ETL cycle are −
Staging Layer − The staging layer is used to store the data extracted from different source data systems.
Data Integration Layer − The integration layer transforms the data from the staging layer and moves the data to a database, where the data is arranged into hierarchical groups, often called dimensions, and into facts and aggregate facts. The combination of facts and dimensions tables in a DW system is called a schema.
Access Layer − The access layer is used by end-users to retrieve the data for analytical reporting.
An ETL tool is used to extract data from different data sources, transform the data, and load it into a DW system. In contrast, a BI tool is used to generate interactive and adhoc reports for end-users, dashboard for senior management, data visualizations for monthly, quarterly, and annual board meetings.
Most common ETL tools include − SAP BO Data Services (BODS), Informatica, Microsoft – SSIS, Oracle Data Integrator ODI, Talend Open Studio, Clover ETL Open source, etc.
Most common BI tools include − SAP Business Objects, SAP Lumira, IBM Cognos, JasperSoft, Microsoft BI Platform, Tableau, Oracle Business Intelligence Enterprise Edition, etc.
The popular ETL tools available in the market are −
Staging area is an intermediate area that sits between data sources and data warehouse/data marts systems. Staging areas can be designed to provide many benefits, but the primary motivations for their use are to increase efficiency of ETL processes, ensure data integrity, and support data quality operations.
Data warehousing is a broader concept as compared to data mining. Data mining involves extracting hidden information from data and interpret it for future predictions. In contrast data warehousing includes operations such as analytical reporting to generate detailed reports and ad-hoc reports, information processing to generate interactive dashboards and charts.
OLTP stands for Online Transactional Processing system which is commonly a relational database and is used to manage day-to-day transactions.
OLAP stands for Online Analytical Processing system which is commonly a multidimensional system and is also called data warehouse.
Suppose a company sells its products to customers. Every sale is a fact that takes place within the company and the fact table is used to record these facts. Each fact table stores the primary keys to join the fact table to dimension tables and measures/facts.
Example − Fact_Units
|Cust_ID||Prod_Id||Time_Id||No. of units sold|
A dimension table stores attributes or dimensions that describe the objects in a fact table. It is a set of companion tables to a fact table.
Example − Dim_Customer
A data mart is a simple form of data warehouse and it is focused on a single functional area. It usually gets data only from a few sources.
Example − In an organization, data marts may exists for Finance, Marketing, Human Resource, and other individual departments which store data related to their specific functions.
Aggregate functions are used to group multiple rows of a single column to form a more significant measurement. They are also used for performance optimization when we save aggregated tables in data warehouse.
Common Aggregate functions are −
|MIN||returns the smallest value in a given column|
|MAX||returns the largest value in a given column|
|SUM||returns the sum of the numeric values in a given column|
|AVG||returns the average value of a given column|
|COUNT||returns the total number of values in a given column|
|COUNT(*)||returns the number of rows in a table|
SELECT AVG(salary) FROM employee WHERE title = 'developer';
Data Definition Language (DDL) statements are used to define the database structure or schema.
CREATE − to create objects in a database
ALTER − alters the structure of a database
Data Manipulation Language (DML) statements are used for manipulate data within database.
SELECT − retrieves data from the a database
INSERT − inserts data into a table
UPDATE − updates existing data within a table
DELETE − deletes all records from a table, the space for the records remain
Data Control Language (DCL) statements are used to control access on database objects.
GRANT − gives user's access privileges to database
REVOKE − withdraws access privileges given with the GRANT command
Operators are used to specify conditions in an SQL statement and to serve as conjunctions for multiple conditions in a statement. The common operator types are −
The common set operators in SQL are −
Intersect operation is used to combine two SELECT statements, but it only returns the records which are common from both SELECT statements. In case of Intersect, the number of columns and datatype must be same. MySQL does not support INTERSECT operator. An Intersect query looks as follows −
select * from First INTERSECT select * from second
Minus operation combines result of two Select statements and return only those result which belongs to first set of result. A Minus query looks as follows −
select * from First MINUS select * from second
If you perform source minus target and target minus source, and if the minus query returns a value, then it should be considered as a case of mismatching rows.
If the minus query returns a value and the count intersect is less than the source count or the target table, then the source and target tables contain duplicate rows.
Group-by clause is used with select statement to collect similar type of data. HAVING is very similar to WHERE except the statements within it are of an aggregate nature.
SELECT dept_no, count ( 1 ) FROM employee GROUP BY dept_no; SELECT dept_no, count ( 1 ) FROM employee GROUP BY dept_no HAVING COUNT( 1 ) > 1;
Example − Employee table
Group by Country
ETL Testing is done before data is moved into a production Data Warehouse system. It is sometimes also called as Table Balancing or production reconciliation.
The main objective of ETL testing is to identify and mitigate data defects and general errors that occur prior to processing of data for analytical reporting.
The following table captures the key features of Database and ETL testing and their comparison −
|Function||Database Testing||ETL Testing|
|Primary Goal||Data validation and Integration||Data Extraction, Transform and Loading for BI Reporting|
|Applicable System||Transactional system where business flow occurs||System containing historical data and not in business flow environment|
|Common Tools in market||QTP, Selenium, etc.||QuerySurge, Informatica, etc.|
|Business Need||It is used to integrate data from multiple applications, Severe impact.||It is used for Analytical Reporting, information and forecasting.|
|Database Type||It is normally used in OLTP systems||It is applied to OLAP systems|
|Data Type||Normalized data with more joins||De-normalized data with less joins, more indexes and Aggregations.|
ETL testing can be divided into the following categories based on their function −
Source to Target Count Testing − It involves matching of count of records in source and target system.
Source to Target Data Testing − It involves data validation between source and target system. It also involves data integration and threshold value check and Duplicate data check in target system.
Data Mapping or Transformation Testing − It confirms the mapping of objects in source and target system. It also involves checking functionality of data in target system.
End-User Testing − It involves generating reports for end users to verify if data in reports are as per expectation. It involves finding deviation in reports and cross check the data in target system for report validation.
Retesting − It involves fixing the bugs and defects in data in target system and running the reports again for data validation.
System Integration Testing − It involves testing all the individual systems, and later combine the result to find if there is any deviation.
Data loss during the ETL process.
Incorrect, incomplete or duplicate data.
DW system contains historical data so data volume is too large and really complex to perform ETL testing in target system.
ETL testers are normally not provided with access to see job schedules in ETL tool. They hardly have access on BI Reporting tools to see final layout of reports and data inside the reports.
Tough to generate and build test cases as data volume is too high and complex.
ETL testers normally doesn’t have an idea of end user report requirements and business flow of the information.
ETL testing involves various complex SQL concepts for data validation in target system.
Sometimes testers are not provided with source to target mapping information.
Unstable testing environment results delay in development and testing the process.
The key responsibilities of an ETL tester include −
Verifying the tables in the source system − Count check, Data type check, keys are not missing, duplicate data.
Applying the transformation logic before loading the data: Data threshold validation, surrogate ky check, etc.
Data Loading from the Staging area to the target system: Aggregate values and calculated measures, key fields are not missing, Count Check in target table, BI report validation, etc.
Testing of ETL tool and its components, Test cases − Create, design and execute test plans, test cases, Test ETL tool and its function, Test DW system, etc.
A transformation is a set of rules which generates, modifies, or passes data. Transformation can be of two types − Active and Passive.
In an active transformation, the number of rows that is created as output can be changed once a transformation has occurred. This does not happen during a passive transformation. The information passes through the same number given to it as input.
Partitioning is when you divide the area of data store in parts. It is normally done to improve the performance of transactions.
If your DW system is huge in size, it will take time to locate the data. Partitioning of storage space allows you to find and analyze the data easier and faster.
Parting can be of two types − round-robin partitioning and Hash partitioning.
In round-robin partitioning, data is evenly distributed among all the partitions so the number of rows in each partition is relatively same. Hash partitioning is when the server uses a hash function in order to create partition keys to group the data.
A Mapplet defines the Transformation rules.
Sessions are defined to instruct the data when it is moved from source to target system.
A Workflow is a set of instructions that instructs the server on task execution.
Mapping is the movement of data from the source to the destination.
Lookup transformation allows you to access data from relational tables which are not defined in mapping documents. It allows you to update slowly changing dimension tables to determine whether the records already exist in the target or not.
A Surrogate key is something having sequence-generated numbers with no meaning, and just to identify the row uniquely. It is not visible to users or application. It is also called as Candidate key.
A Surrogate key has sequence-generated numbers with no meaning. It is meant to identify the rows uniquely.
A Primary key is used to identify the rows uniquely. It is visible to users and can be changed as per requirement.
In such cases, you can apply the checksum method. You can start by checking the number of records in the source and the target systems. Select the sums and compare the information.
In this testing, a tester validates the range of data. All the threshold values in the target system are to be checked to ensure they are as per the expected result.
Example − Age attribute shouldn’t have a value greater than 100. In Date column DD/MM/YY, month field shouldn’t have a value greater than 12.
Select Cust_Id, Cust_NAME, Quantity, COUNT (*) FROM Customer GROUP BY Cust_Id, Cust_NAME, Quantity HAVING COUNT (*) >1;
When no primary key is defined, then duplicate values may appear.
Data duplication may also arise due to incorrect mapping, and manual errors while transferring data from source to target system.
Regression testing is when we make changes to data transformation and aggregation rules to add a new functionality and help the tester to find new errors. The bugs that appear in data which comes in Regression testing are called Regression.
The three approaches are − top-down, bottom-up, and hybrid.
The most common ETL testing scenarios are −
Data purging is a process of deleting data from a data warehouse. It removes junk data like rows with null values or extra spaces.
Cosmetic bug is related to the GUI of an application. It can be related to font style, font size, colors, alignment, spelling mistakes, navigation, etc.
It is called Boundary Value Analysis related bug.
You can do it by creating a mapping variable and a filtered transformation. You might need to generate a sequence in order to have the specifically sorted record you require.
Value comparison − It involves comparing the data in the source and the target systems with minimum or no transformation. It can be done using various ETL Testing tools such as Source Qualifier Transformation in Informatica.
Critical data columns can be checked by comparing distinct values in source and target systems.
You can use Minus and Intersect statements to perform data completeness validation. When you perform source minus target and target minus source and the minus query returns a value, then it is a sign of mismatching rows.
If the minus query returns a value and the count intersect is less than the source count or the target table, then duplicate rows exist.
Shortcut Transformation is a reference to an object that is available in a shared folder. These references are commonly used for various sources and targets which are to be shared between different projects or environments.
In the Repository Manager, a shortcut is created by assigning ‘Shared’ status. Later, objects can be dragged from this folder to another folder. This process allows a single point of control for the object and multiple projects do not have all import sources and targets into their local folders.
Reusable Transformation is local to a folder. Example − Reusable sequence generator for allocating warehouse Customer ids. It is useful to load customer details from multiple source systems and allocating unique ids to each new source-key.
When you join a single table to itself, it is called Self-Join.
Database normalization is the process of organizing the attributes and tables of a relational database to minimize data redundancy.
Normalization involves decomposing a table into less redundant (and smaller) tables but without losing information.
A fact-less fact table is a fact table that does not have any measures. It is essentially an intersection of dimensions. There are two types of fact-less tables: One is for capturing an event, and the other is for describing conditions.
Slowly Changing Dimensions refer to the changing value of an attribute over time. SCDs are of three types − Type 1, Type 2, and Type 3.