Data Quality

Introduction

The data quality task is used to check the data accuracy during the integration and processing of data. Data quality tasks in this release include single-table checking, single-table custom SQL checking, multi-table accuracy, and two-table value comparisons. The running environment of the data quality task is Spark 2.4.0, and other versions have not been verified, and users can verify by themselves.

The execution logic of the data quality task is as follows:

• The user defines the task in the interface, and the user input value is stored in TaskParam.
• When running a task, Master will parse TaskParam, encapsulate the parameters required by DataQualityTask and send it to Worker.
• Worker runs the data quality task. After the data quality task finishes running, it writes the statistical results to the specified storage engine.
• The current data quality task result is stored in the t_ds_dq_execute_result table of dolphinscheduler Worker sends the task result to Master, after Master receives TaskResponse, it will judge whether the task type is DataQualityTask, if so, it will read the corresponding result from t_ds_dq_execute_result according to taskInstanceId, and then The result is judged according to the check mode, operator and threshold configured by the user.
• If the result is a failure, the corresponding operation, alarm or interruption will be performed according to the failure policy configured by the user.
• Add config : <server-name>/conf/common.properties

# Change to specific version if you not use dev branch
data-quality.jar.name=dolphinscheduler-data-quality-dev-SNAPSHOT.jar

• Please fill in data-quality.jar.name according to the actual package name.
• If you package data-quality separately, remember to modify the package name to be consistent with data-quality.jar.name.
• If the old version is upgraded and used, you need to execute the sql update script to initialize the database before running.
• If you want to use MySQL data, you need to comment out the scope of MySQL in pom.xml.
• Currently only MySQL, PostgreSQL and HIVE data sources have been tested, other data sources have not been tested yet.
• Spark needs to be configured to read Hive metadata, Spark does not use jdbc to read Hive.

Detailed Inspection Logic

In the example, assuming that the actual value is 10, the operator is >, and the expected value is 9, then the result 10 -9 > 0 is true, which means that the row data in the empty column has exceeded the threshold, and the task is judged to fail.

Task Operation Guide

Null Value Check for Single Table Check

Inspection Introduction

The goal of the null value check is to check the number of empty rows in the specified column. The number of empty rows can be compared with the total number of rows or a specified threshold. If it is greater than a certain threshold, it will be judged as failure.

• The SQL statement that calculates the null of the specified column is as follows:

  SELECT COUNT(*) AS miss FROM ${src_table} WHERE (${src_field} is null or ${src_field} = '') AND (${src_filter})

• The SQL to calculate the total number of rows in the table is as follows:

  SELECT COUNT(*) AS total FROM ${src_table} WHERE (${src_filter})

Interface Operation Guide

Timeliness Check of Single Table Check

Inspection Introduction

The timeliness check is used to check whether the data is processed within the expected time. The start time and end time can be specified to define the time range. If the amount of data within the time range does not reach the set threshold, the check task will be judged as fail.

Interface Operation Guide

Field Length Check for Single Table Check

Inspection Introduction

The goal of field length verification is to check whether the length of the selected field meets the expectations. If there is data that does not meet the requirements, and the number of rows exceeds the threshold, the task will be judged to fail.

Interface Operation Guide

Uniqueness Check for Single Table Check

Inspection Introduction

The goal of the uniqueness check is to check whether the fields are duplicated. It is generally used to check whether the primary key is duplicated. If there are duplicates and the threshold is reached, the check task will be judged to be failed.

Interface Operation Guide

Regular Expression Check for Single Table Check

Inspection Introduction

The goal of regular expression verification is to check whether the format of the value of a field meets the requirements, such as time format, email format, ID card format, etc. If there is data that does not meet the format and exceeds the threshold, the task will be judged as failed.

Interface Operation Guide

Enumeration Value Validation for Single Table Check

Inspection Introduction

The goal of enumeration value verification is to check whether the value of a field is within the range of the enumeration value. If there is data that is not in the range of the enumeration value and exceeds the threshold, the task will be judged to fail.

Interface Operation Guide

Table Row Number Verification for Single Table Check

Inspection Introduction

The goal of table row number verification is to check whether the number of rows in the table reaches the expected value. If the number of rows does not meet the standard, the task will be judged as failed.

Interface Operation Guide

Custom SQL Check for Single Table Check

Interface Operation Guide

Accuracy Check of Multi-table

Inspection Introduction

Accuracy checks are performed by comparing the accuracy differences of data records for selected fields between two tables, examples are as follows

• table test1

• table test2

If you compare the data in c1 and c21, the tables test1 and test2 are exactly the same. If you compare c2 and c22, the data in table test1 and table test2 are inconsistent.

Interface Operation Guide

Comparison of the values checked by the two tables

Inspection Introduction

Two-table value comparison allows users to customize different SQL statistics for two tables and compare the corresponding values. For example, for the source table A, the total amount of a certain column is calculated, and for the target table, the total amount of a certain column is calculated. value sum2, compare sum1 and sum2 to determine the check result.

Interface Operation Guide

Task result view

Rule View

List of rules

Rules Details