Horizontal and Vertical Partitioning in Replication

By Guest Authors on 30 January 2003 | Tags: Replication


This article was written by Lynn Zhu and Demico Quinn. They write "Before coming up with the solution that this article will describe, we decided that this would be a great topic to write about and share our experiences with others. We also found that there were not a lot of articles out there that focused on Replication. More specifically, articles that chronicled case studies where column and row filters were being used as an integral part of the solution. It is our hope that you will find this article to be helpful and inspire readers to write more articles about Replication."

Using Replication’s Horizontal and Vertical partitioning capabilities to manage pubications in a distributed database environment

There are times that you may find yourself on a project where you have one data source and need to publish to many subscribing databases. Often times, Replication is not looked at as a viable solution to achieving this goal. We can only attribute this to the “blood sweat and tears” that it took to get Replication in SQL 6.5 to work smoothly. Fortunately those days are gone and Microsoft has only improved the way Replication works in SQL 2000. For this project, we will demonstrate the way we utilized the Horizontal and Vertical partitioning capabilities that Microsoft has added to the Replication utility.

To begin, we have company A, which sells it’s products based on district as the following scheme illustrates:

In this scenario a products given retail price may differ depending upon whether the product was purchased using the website or sold at one of our store locations. Additionally, a product can also have different discount prices in different districts, and be sold on either the web or in the store. Often times the discounts that are found on the website may or may not be applicable to “in store sales”, and the same rule applies to discounts found in the store.

All transactions will write to one database, which will act as the publisher to multiple subscribing databases. The database is called CentralInfo, and as earlier stated will act as the publishing database for our demonstration. There will be two databases that will subscribe to publications that are based off of the CentralInfo database. The WebSale database only needs the information about the products, web sale prices, and discounts that will be sold through the web by districts. The StoreSale database only needs the information about the products, store sale prices, and discounts that will be sold through the store by districts. To meet this business requirement we had to make some small modifications to the scheme. In order to restrict the rows and columns that will be defined in the publications, we added two columns (Use4WebSale and Use4StoreSale) to the Products and Products2District tables as row filters.

(After adding Use4WebSale, and Use4StoreSale columns)

The filters are defined as integer data types and the default value for columns Use4WebSale and Use4StoreSale is 0. If a given product is for web sales then the value of column Use4WebSale will be updated to 1. If a given product is for store sales then the value of column Use4StoreSale will be updated to 1. For products that are meant for both web and store sales the value of both Use4WebSale and Use4StoreSale are set to 1. The following chart illustrates the possible scenarios.

Product2Distirct:

productid districtid websaleprice storesaleprice discountid discountstartdate discountenddate use4websale use4storesale
3 2 20 15 1 12/16/2002 7/31/2003 0 1
4 2 30 25 2 12/25/2002 1/15/2003 1 1
5 2 25 20 1 12/16/2002 7/31/2003 1 0

Products:

productid productname productmodel warehouseid producttypeid use4websale productcost use4storesale
3 Oil filter RS-300 2 1 1 $7.99 1
4 Side mirror PA-100 2 2 0 $19.99 1
5 Door Bulb DB-022 1 3 1 $15.99 0
6 Window Bulb BW-012 1 3 0 $6.99 0

Publications and Subscribers

There will be two publications and two subscribers to be set up.

1a. Store publication configurations

Now that we have established what changes needed to be made to the scheme, we can now start defining our publications. We will define store publication and identify the articles that will be used in the publication. To create the publication the “Create Publication wizard” was used. And defined as follows:

Publication Name: CentralInfo_to_StoreSale

Publishing Database: CentralInfo

Articles:

Products
Products2Districts
Districts
States
WareHouses
Discounts
ProductType

Snapshot Options : The following options refer to the initial snapshot.

  • Drop existing tables and recreate
  • Uncheck “Include declared referential integrity”
  • Uncheck Clustered indexes
  • Uncheck Non-clustered indexes

After initial snapshot has been applied, the filters can be added to the articles and re-initialization occurs on the publication. You must also change the snapshot property to “ Delete data in the existing tables that match the row filter statement”. In the case where you have populated your CentralInfo database with data, you can apply all filters prior to the initial snapshot. This will eliminate the need to run the initialization snapshot a second time. Filters are defined as follows:

For the Products article add this row filter (table) Row Filter: Use4StoreSale =1

Syntax: SELECT  <published_columns>
             FROM  <<TABLE>>
             WHERE   <TABLE>>.Use4StoreSale = 1

For the Products2Districts article add this row filter(table) Row Filter and Column Filter: Use4StoreSale =1

Under the Column filter tab for the article “Products2Districts”, uncheck column WebSalePrice

Syntax: SELECT  <published_columns>
             FROM  <<TABLE>>
             WHERE  <<TABLE>>.Use4StoreSale = 1

1b. Store subscriber configurations

The following describes the subscriber options that define how the StoreSale database will receive data from CentralInfo Database. The subscribing database can be created during the subscription process or pre-exist.

Subscription name: Server name: StoreSale
Type: Push
Publish Interval: Optional
Scheme: Scheme is created by the Snapshot agent

Subscriber StoreSale:

2a. Web publication configurations

Now, we will define the web publication and identify the articles that will be used. To create the publication the “Create Publication wizard” was used. And defined as follows:

Publication Name: CentralInfo_to_WebSale

Publishing Database: CentralInfo

Articles:

Products
Products2Districts
Districts
States
WareHouses
Discounts
ProductType

Note: The articles in this publication are the same as defined in our StoreSale publication. Repeat all steps in this publication as you did for the StoreSale publication, the only exception will be to filter on the column; “Use4WebSale”.

2b. Web subscriber configurations

The following describes the subscriber options that define how the WebSale database will receive data from CentralInfo Database. The subscribing database can be created during the subscription process or pre-exist.

Name: SQLSERVERNAME: WebSale
Type: Push
Publish Interval: Optional
Scheme: Scheme is created by the Snapshot agent

Summary of implementation steps

The following is a step-by-step summary of the tasks performed in order to implement this solution.

1.	Run script to create the database scheme (CentralInfo). 
2.	Create Transactional Publication which includes the following sub tasks:
2.1	Create StoreSale database
2.2	Define articles with article options
2.3	No article filters
3.	Run Snapshot
4.	Run distribution
5.	Modify Transactional publication which includes the following sub tasks:
5.1	Add filter to each article with the following  “Use4StoreSale =1”
5.2	Update Snapshot option to “ Delete data in the existing tables 
        that match the row filter statement”
5.3	Re-initialize the publication
6.	Create Transactional Publication which includes the following sub tasks:
6.1	Create WebSale database
6.2	Define articles with article options
6.3	No article filters
7.	Run Snapshot
8.	Run Distribution
9.	Modify Transactional publication which includes the following sub tasks:
9.1	Add filter to each article with the following “Use4WebSale =1”
9.2	Update Snapshot option to “ Delete data in the existing 
        tables that match the row filter statement”
10.	Re-initialize the publication
11.	Run Snapshot (Once data has been mapped via the Customer Care department)
12.	Run Distribution
13.	Run scripts to load data into the CentralInfo database.
14.	Once data has been published to the subscribers, check 
        to insure that the filters are working as expected.

Lynn Zhu is the Lead Developer of Western Wireless Corporation and is responsible for SQL Server database design and development. She is an MCDBA. Demico Quinn is the SQL Server DBA for the Western Wireless Corporation where he is responsible for SQL Server database servers and database applications.


- Advertisement -