QA Guidelines for Data Warehouse Quality Verification

QA Guidelines for Data Warehouse Quality Verification

This document describes testing guidelines and steps for verifying data, ETL processes, and SQL during the construction, unit testing, system and integration testing of an application’s data warehouse operational tables and data mart.
1.) Verify and Maintain the Data Low Level Design (LLD)
A first level of testing and validation begins with the formal acceptance of the logical data model and “low level design” (LLD). All further testing and validation will be based on the understanding of each of the data elements in the model.
Data elements that are created through a transformation or summary process must be clearly identified and calculations for each of these data elements must be clear and easily interpreted.
During the LLD reviews and updates, special consideration should be given to typical modeling scenarios that exist in the project. Examples follow:
1. Verify that many-to-many attribute relationships are clarified and resolved.
2. Verify the types of keys that are used: surrogate keys versus natural keys.
3. Verify that the business analyst / DBA reviewed with ETL architect and developers (application) the lineage and business rules for extracting, transforming, and loading the data warehouse?
4. Verify that all transformation rules, summarization rules, and matching and consolidation rules have clear specifications.
5. Verify that specified transformations, business rules and cleansing specified in LLD and other application logic specs have been coded correctly in ETL, JAVA, and SQL used for data loads.
6. Verify that procedures are documented to monitor and control data extraction, transformation and loading. The procedures should describe how to handle exceptions and program failures.
7. Verify that data consolidation of duplicate or merged data was properly handled.
8. Verify that samplings of domain transformations will be taken to verify they are properly changed.
9. Compare unique values of key fields between source data and data loaded to the warehouse. This is a useful technique that points out a variety of possible data errors without doing a full validation on all fields.
10. Validate that target data types are as specified in the design and/or the data model.
11. Verify how sub-class/super-class attributes depicted?
12. Verify that data field types and formats are specified.
13. Verify that defaults are specified for fields where needed.
14. Verify that processing for invalid field values in source are defined
15. Verify that expected ranges of field contents are specified where known.
16. Verify that keys generated by the “sequence generator” are identified.
17. Verify that slowly changing dimensions are described?

2.) Analyze Source Data Before & After Extraction to Staging

Testers should extract representative data from each source file (before or after extract to staging tables) and confirm that the data is consistent with its definition; QA can discover any anomalies in how the data is represented and write defect reports where necessary. The objective is to discover data that does not meet “data quality factors” as described in specifications. See list below and Table 1.
This verification process will be used for temp tables used in a step process for data transformations, cleaning, etc.
• Verify that the scope of values in each column are within specifications
• Identify unexpected values in each field
• Verify relationships between fields
• Identify frequencies of values in columns and whether these frequencies make sense?

Inputs: Application source data models and low level data design, data dictionaries, data attribute sources.
Outputs: Newly discovered attributes, undefined business rules, data anomalies such as fields used for multiple purposes.
Techniques and Tools: Data extraction software, business rule discovery software, data analysis tools.
Process Description:
1. Extract representative samples of data from each source or staging table.
2. Parse the data for the purpose of profiling.
3. Verify that not-null fields are populated and populated as expected.
4. Structure discovery – Does the data match the corresponding metadata? Do field attributes of the data match expected patterns? Does the data adhere to appropriate uniqueness and null value rules?
5. Data discovery – Are the data values complete, accurate and unambiguous?
6. Relationship discovery – Does the data adhere to specified required key relationships across columns and tables? Are there inferred relationships across columns, tables or databases? Is there redundant data?
7. Verify that all required data from the source was extracted. Verify that extraction process did not extract more or less data source than it should have.
8. Verify or write defects for exceptions and errors discovered during the ETL process.
9. Verify that extraction process did not extract duplicate data from the source (usually this happens in repeatable processes where at point zero we need to extract all data from the source file, but the during the next intervals we only need to capture the modified, and new rows.).
10. Validate that no data truncation occurred during staging.
11. Utilize a data profiling tool or methods that show the range and value distributions of fields in the source data. This is used to identify any data anomalies from source systems that may be missed even when the data movement is correct.
12. Validation & Certification Method: it is sufficient to identify the requirements and count (via SQL) the number of rows that should be extracted from the source systems. The QA team will also count the number of rows in the result / target sets and match the two for validation. The QA team will maintain a set of SQL statements that are automatically run at this stage to validate that no duplicate data have been extracted from the source systems.
Table 1: Data Quality Factors
FACTOR DESCRIPTION EXAMPLE
Data Consistency Issues:
Varying Data Definitions The data type and length for a particular attribute may vary in files or tables though the semantic definition is the same. Account number may be defined as: Number (9) in one field or table and Varchar2(11) in another table
Misuse of Integrity Constraints When referential integrity constraints are misused, foreign key values may be left “dangling” or inadvertently deleted. An account record is missing but dependent records are not deleted.
Nulls Nulls when field defined as “not-null”. The company has been entered as a null value for a business. A report of all companies would not list the business.
Data Completeness Issues:
Missing data Data elements are missing due to a lack of integrity constraints or nulls that are inadvertently not updated. An account date of estimated arrival is null thus impacting an assessment of variances in estimated/actual account data.
Inaccessible Data Inaccessible records due to missing or redundant identifier values. Business numbers are used to identify a customer record. Because uniqueness was not enforced, the business ID (45656) identifies more than one customer.
Missing Integrity Constraints Missing constraints can cause data errors due to nulls, non-uniqueness, or missing relationships. Account records with a business identifier exist in the database but cannot be matched to an existing business.
Data Correctness Issues:
Loss Projection Tables that are joined over non key attributes will produce non existent data that is shown to the user. Lisa Evans works in the LA office in the Accounting department. When a report is generated, it shows her working in IT department.
Incorrect Data Values Data that is misspelled or inaccurately recorded. 123 Maple Street is recorded with a spelling mistake and a street abbreviation (123 Maple St)
Inappropriate Use of Views Data is updated incorrectly through views. A view contains non key attributes from base tables. When the view is used to update the database, null values are entered into the key columns of the base tables.
Disabled Integrity Constraints Null, non unique, or out of range data may be stored when the integrity constraints are disabled. The primary key constraint is disabled during an import function. Data is entered into the existing data with null unique identifiers.
Non-duplication Testing should be conducted to determine if there’s duplication of data where there should not be. Duplicate rows or column data.
Misuse of Integrity Constraints Check whether null or foreign key constraints are inappropriate or too restrictive. Check constraint only allows hard coded values of “C”, “A”, “X”, and “Z”. But a new code “B” cannot be entered.
Data Comprehension Issues:
Data Aggregation Aggregated data is used to represent a set of data elements. One name field is used to store surname, first name, middle initial, and last name (e.g., John, Hanson, Mr.).
Cryptic Object Definitions Database object (e.g., column) has a cryptic, unidentifiable name. Customer table with a column labeled, “c_avd”. There is no documentation as to what the column might contain.
Unknown or Cryptic Data Cryptic data stored as codes, abbreviations, truncated, or with no apparent meaning. Shipping codes used to represent various parts of the customer base (‘01’, ‘02’, ‘03’). No supporting document to explain the meaning of the codes.
Accuracy Data will be matched against business rules. Boundary values (low, high’s) will be identified for relevant fields and compared with expectations.
Completeness Data will be assessed to verify that all required is present. Missing rows will be identified; Null values will be identified in data elements where a value is expected.
Precision Precision testing is conducted to evaluate the level of data not sufficiently precise based on specifications.

3.) Verify Corrected, Cleaned, Source Data in Staging

This step works to improve the quality of existing data in source files or “defects” that meet source specs but must be corrected before load.
Inputs:
 Files or tables (staging) that require cleansing; data definition and business rule documents, data map of source files and fields; business rules, data anomalies discovered in earlier steps of this process.
 Fixes for data defects that will result in data that does not meet specifications for the application DW.
Outputs: Defect reports, cleansed data, rejected or uncorrectable data
Techniques and Tools: Data reengineering, transformation, and cleansing tools, MS Access, Excel filtering.
Process Description: In this step, data with missing values, known errors, and suspect data is corrected. Automated tools may be identified to best to locate, clean / correct large volumes of data.
1. Document the type of data cleansing approach taken for each data type in the repository.
2. Determine how “uncorrectable” or suspect data is processed, rejected, maintained for corrective action. SME’s and stakeholders should be involved in the decision.
3. Review ETL defect reports to assess rejected data excluded from source files or information group targeted for the warehouse.
4. Determine if data not meeting quality rules was accepted.
5. Document in defect reports, records and important fields that cannot be easily corrected.
6. Document records that were corrected and how corrected.
Certification Method: Validation of data cleansing processes could be a tricky proposition, but certainly doable. All data cleansing requirements should be clearly identified. The QA team should learn all data cleansing tools available and their methods. QA should create various conditions as specified in the requirements for the data cleansing tool to support and validate its results. QA will run a volume of real data through each tool to validate accuracy as well as performance.

4.) Verifying Matched and Consolidated Data

There are often ETL processes where data has been consolidated from various files into a single occurrence of records. The cleaned and consolidated data can be assessed to very matched and consolidated data.
Much of the ETL heavy lifting occurs in the transform step where combined data, data with quality issues, updated data, surrogate keys, build aggregates, are processed.
Inputs: Analysis of all files or databases for each entity type
Outputs:
Report of matched, consolidated, related data that is suspect or in error
List of duplicate data records or fields
List of duplicate data suspects.
Techniques and Tools: Data matching techniques or tools; data cleansing software with matching and merging capabilities.
Process Description:
1. Establish match criteria for data. Select attributes to become the basis for possible duplicate occurrences (e.g., names, account numbers).
2. Determine the impact of incorrectly consolidated records. If the negative impact of consolidating two different occurrences such as different customers into a single customer record exists, submit defect reports. The fix should be higher controls to help avoid such consolidations in the future.
3. Determine the matching techniques to be used: Exact character match in two corresponding fields such as wild card match, key words, close match, etc.
4. Compare match criteria for specific record with all other records within a given file to look for intra-file duplicate records.
5. Compare match criteria for a specific record with all records in another file to seek inter-file duplicate records.
6. Evaluate potential matched occurrences to assure they are, in fact, duplicate.
7. Verify that consolidated data into single occurrences is correct.
8. Examine and re-relate data related to old records being consolidated to new occurrence-of-reference record. Validate that no related data was overlooked.

5.) Verify Transformed / Enhanced / Calculated Data to Target Tables

At this stage, base data is being prepared for loading into the Application operational tables and the data mart. This includes converting and formatting cleansed, consolidated data into the new data architecture and possibly enhancing internal operational data with external data licensed from service providers.
The objective is to successfully map the cleaned, corrected and consolidated data into the DW environment.
Inputs: Cleansed, consolidated data; external data from service providers; business rules governing the source data; business rules governing the target DW data; transformation rules governing the transformation process; DW or target data architecture; data map of source data to standardized data.
Output: Transformed, calculated, enhanced data; updated data map of source data to standardized data; data map of source data to target data architecture
Techniques and Tools: Data transformation software; external or online or public databases.
Process Description:
1. Verify that the data warehouse construction team is using the data map of source data to the DW standardized data, verify the mapping.
2. Verify that the data transformation rules and routines are correct.
3. Verify the data transformations to the DW and assure that the processes were performed according to specifications.
4. Verify that data loaded in the operational tables and data mart meets the definition of the data architecture including data types, formats, accuracy, etc.
5. Develop scenarios to be covered in Load Integration Testing
6. Count Validation: Record Count Verification DWH backend/Reporting queries against source and target as an initial check.
7. Dimensional Analysis: Data integrity exists between the various source tables and parent / child relationships.
8. Statistical Analysis: Validation for various calculations.
9. Data Quality Validation: - Check for missing data, negatives and consistency. Field-by-field data verification will be done to check the consistency of source and target data.
10. Granularity: Validate at the lowest granular level possible (lowest in the hierarchy E.g. Country-City-Sector– start with test cases).
11. Dynamic Transformation Rules & Tables: such methods need to be checked continuously to ensure the correct transformation routines are executed. Verify that dynamic mapping tables and dynamic mapping rules provide an easy, documented, and automated way for transforming values from one or more sources into a standard value presented in the DW.
12. Verification Method: The QA team will identify the detailed requirements as they relate to transformation and validate the dynamic transformation rules and tables against DW records. Utilizing SQL and related tools, the team will identify unique values in source data files that are subject to transformation. The QA team identifies the results from the transformation process and validate that such transformation have accurately taken place.

6.) Front-end UI and Report Testing Using Operational Tables and Data Mart

End user reporting is a major component of the Application Project. The report code may run aggregate SQL queries against the data stored in the data mart and/or the operational tables then display results in a suitable format either in a Web browser or on a client application interface. Once the initial view is rendered, the reporting tool interface provides various ways of manipulating the information such as sorting, pivoting, computing subtotals, and adding view filters to slice-and-dice the information further. Special considerations such as those below will be prepared while testing the reports:
1. The ETL process should be complete, the data mart must be populated and data quality testing should be largely completed.
2. The front-end will use a SQL engine which will generate the SQL based on the how the dimension and fact tables are mapped. Additionally, there may be global or report-specific parameters set to handle very large database (VLDB)-related optimization requirements. As such, testing of the front-end will concentrate on validating the SQL generated; this in turn validates the dimensional model and the report specification vis-à-vis the design.
3. Unit testing of the reports will be conducted to verify the layout format per the design mockup, style sheets, prompts and filters, attributes and metrics on the report.
4. Unit testing will be executed both in the desktop and Web environment.
5. System testing of the reports will concentrate on various report manipulation techniques like the drilling, sorting and export functions of the reports in the Web environment.
6. Reports and/or documents need special consideration for testing because they are high visibility reports used by the top analysts and because they have various charts, gauges and data points to provide a visual insight to the performance of the organization in question.
7. There may be some trending reports, or more specifically called comp reports, that compare the performance of an organizational unit over multiple time periods. Testing these reports needs special consideration especially if a fiscal calendar is used instead of an English calendar for time period comparison.
8. For reports containing derived metrics special focus should be paid to any subtotals. The subtotal row should use a "smart-total," i.e., do the aggregation first and then do the division instead of adding up the individual cost per click of each row in the report.
9. Reports with "non-aggregate-able" metrics (e.g., inventory at hand) also need special attention to the subtotal row. It should not, for example, add up the inventory for each week and show the inventory of the month.
10. During unit testing, all data formats will be verified against the standard. For example, metrics with monetary value should show the proper currency symbol, decimal point precision (at least two places) and the appropriate positive or negative. For example, negative numbers should be shown in red and enclosed in braces.
11. During system testing, while testing the drill-down capability of reports, care will be taken to verify that the subtotal at the drill-down report matches with the corresponding row of the summary report. At times, it is desirable to carry the parent attribute to the drill-down report; verify the requirements for this.
12. When testing reports containing conditional metrics, care will be taken to check for "outer join condition;" i.e., nonexistence of one condition is reflected appropriately with the existence of the other condition.
13. Reports with multilevel sorting will get special attention for testing especially if the multilevel sorting includes both attributes and metrics to be sorted.
14. Reports containing metrics at different dimensionality and with percent-to-total metrics and/or cumulative metrics needs will get special attention to check that the subtotals are hierarchy-aware (i.e., they "break" or "re-initialized" at the appropriate levels).

7.) Operational Table and Data Mart: Build Sanity Test
1. Session Completions: All workflow sessions completed successfully using the Log Viewer.
2. Source to Target Counts: This process verifies that the number of records in the source system matches the number of records received, and ultimately processed, into the data warehouse. If Look-up’s are involved in the ETL process, the count between source and target will not match. The ETL Session log and target table counts are compared.
3. Source to Target Data Verification: The process verifies that all source and reference tables have data before running ETLs. We verify that all target tables were truncated before the load unless target tables are updated. This process verifies that the source field threshold is not subject to truncation during the transformation or loading of data.
4. Field to Field Verification: This process verifies the field values from the source system to target. This process ensures that the data mapping from the source system to the target is correct, and that data sent has been loaded accurately.
5. ETL Exception Processing: Exception processing verification looks for serious data errors that would cause system processing failures or data corruption. An Exception report verifying the number and types of errors encountered is produced and reviewed for additional processing and / or reporting to the customer.
There are two types of Exception process:
1. Database Exception:
• Not Null - Source column is null while target is not null
• Reference Key - The records coming from the source data do not have a corresponding parent key in the parent table.
• Unique Key - The record already exists in the target table.
• Check Constraint - CHECK constraints enforce domain integrity by limiting the values that are accepted by a column
2. Business Exception
These are the exceptions thrown based on certain business rules defined for specific data elements or group of data elements
• ETL process utilizes a single Exception Table to capture the exceptions from various ETL sessions and an Error Lookup table which has various error codes and their description.
• We check the Exception process using the Session Log and Exception Table.

8.) Sanity Test: Exit and Suspension Criteria
1. No critical defects unfixed; No more than 3 high severity defects.
2. 80% or more of build functionality can be tested – functionality might fail because of JAVA / report code.
3. Platform performance is such that test team can productively work to schedule
4. Fewer than 15% of build fixes failed