Hướng dẫn học Microsoft SQL Server 2008 part 38 docx

Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 332 Part II Manipulating Data with Select Creating views using the Query Designer Because a view is nothing more than a saved SQL SELECT statement, the creation of a view begins with aworking SELECT statement. Any SQL SELECT statement, as long as it’s a valid SQL SELECT state- ment (with a few minor exceptions), can be cut and pasted from nearly any other tool into a view. Within SQL Server Management Studio, views are listed in their own node under each database. The New View command in the context menu launches the Query Designer in a mode that creates views, as shown in Figure 14-1. FIGURE 14-1 Creating a view in Management Studio’s Query Designer The View Designer mode functions within Management Studio’s Query Designer, which is also used to query tables. The actual SQL code for the view is displayed or edited in the SQL pane. Columns may be added to the view by using the Diagram pane, the Grid pane, or the SQL pane. The Add Table feature, 332 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 333 Projecting Data Through Views 14 available in the context menu or toolbar, can add tables, other views, synonyms, and table-valued functions. Tables or other views can be added to the new view by dragging them to the Diagram pane from the Object Explorer or using the Add Table context menu option. There’s a toolbar button and a context menu item to add a derived table to the view, but all it does is slightly modify the from clause to create a placeholder for the subquery. The SQL for the subquery is then manually entered in the SQL pane. The Verify SQL Syntax button in the toolbar verifies only the SQL syntax; it does not verify the names of tables, views, or columns in the SQL SELECT statement. To test the view’s SQL SELECT statement within Query Designer, use the Execute SQL button or F5. This will run the SELECT statement by itself, without creating the view. The Save toolbar button actually runs the script to create the view in the database. Note that the view must be a valid, error-free SQL SELECT statement in order to be saved. For more details on using the Query Designer, refer to Chapter 6, ‘‘Using Management Studio.’’ Once the view is created, several tasks may be performed on the view using Object Explorer’s view con- text menu: ■ Redesign the view: Opens the Query Designer tool with the view’s SELECT statement. ■ Selecttopnrows:Opens the Query Editor with a SELECT statement referencing the view. The number of rows selected can be modified in Management Studio’s options. ■ Edit top n rows: Opens the Query Designer with a SELECT statement referencing the view, with only the results pane visible, and executes the view. ■ Script the view: Management Studio can script the DDL statements to CREATE, ALTER,or DROP the view, as well as sample DML statements referencing the view. ■ View dependencies: This option can be very important because views, by definition, reference other data sources, and are often referenced themselves. ■ Full-text indexes: A full-text index (covered in Chapter 19, ‘‘Using Integrated Full-Text Search’’) can be created and managed based on data selected by the view. ■ Policies: Apply and manage policy-based management policies for the view. ■ Rename/Delete the view: The view may also be renamed or dropped by selecting it and pressing Rename or Delete, respectively. ■ Properties: Opens the properties dialog with pages for security permissions and extended properties. Double-clicking the view opens its subnodes: columns, triggers (instead of tasks), indexes (indexed views), and statistics. 333 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 334 Part II Manipulating Data with Select Creating views with DDL code Views may be managed using the Query Editor by executing SQL scripts with the data definition language (DDL) commands: CREATE, ALTER,andDROP. The basic syntax for creating a view is as follows: CREATE VIEW schemaname.ViewName [(Column aliases)] AS SQL Select Statement; For example, to create the view vEmployeeList in code, the following command would be executed in a query window: USE AdventureWorks2008 Go CREATE VIEW dbo.vEmployeeList AS SELECT P.BusinessEntityID, P.Title, P.LastName, P.FirstName, E.JobTitle FROM Person.Person P INNER JOIN HumanResources.Employee E ON P.BusinessEntityID = E.BusinessEntityID As with creating any object, the create command must be the only command in the batch. Although I’m generally opposed to Hungarian notation (tblTablename, intIntegerColumn, etc.) for database objects, I prefer to preface views with a lowercase v, simply to keep them separate in data source listings, but, to be honest, most database developers do not preface views with a v. The view name must be unique in the database. Attempting to create a view with a name shared by any other object will generate an error. Executing views Technically, a view by itself cannot be executed. A view can only patiently wait to be referenced by a SQL query. Aquery( SELECT, INSERT, UPDATE, DELETE,orMERGE) can include the view as a data source, and that query can be executed. As illustrated in Figure 14-2, a view is useful only as a data source within a query. You can think of a view as nothing more than a placeholder for a saved SELECT statement. The following SELECT statement references the vEmployeeList view: SELECT BusinessEntityID, LastName, FirstName, JobTitle FROM dbo.vEmployeeList 334 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 335 Projecting Data Through Views 14 FIGURE 14-2 When the query that references a view is submitted to SQL Server, the query parser picks the query apart and replaces the name of the view with the view’s select statement. ad hoc Query VIEW dbo.vEmployeeList SELECT BusinessEntityID, LastName, FirstName, JobTitle FROM dbo.vEmployeeList; SELECT P.BusinessEntityID, P.LastName, P.FirstName, E.JobTitle FROM Person.Person P INNERJOIN HumanResources.Employee E ON P.BusinessEntityID = E.BusinessEntityID Query Parser includes the view into the submitted query Executed Query SELECT BusinessEntityID, LastName, FirstName, JobTitle FROM (SELECT P.BusinessEntityID, P.LastName, P.FirstName, E.JobTitle FROM Person.Person P INNER JOIN HumanResources.Employee E ON P.BusinessEntityID = E.BusinessEntityID); Result (abbreviated): BusinessEntityID LastName FirstName JobTitle 1S ´ anchez Ken Chief Executive Officer 2 Duffy Terri Vice President of Engineering 3 Tamburello Roberto Engineering Manager 4 Walters Rob Senior Tool Designer When views are referenced from ad hoc queries, a WHERE condition is typically added to filter the data from the view: SELECT BusinessEntityID, LastName, FirstName, JobTitle FROM dbo.vEmployeeList WHERE JobTitle = ‘Database Administrator’; 335 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 336 Part II Manipulating Data with Select Result: BusinessEntityID LastName FirstName JobTitle 270 Ajenstat Fran ¸ coi Database Administrator 271 Wilson Dan Database Administrator Altering and dropping a view It’s likely that the view’s SELECT statement will need to be changed at some point in time. Once a view has been created, the SQL SELECT statement may be easily edited by using the ALTER command. Alter- ing the view changes the saved SELECT statement while keeping any properties and security settings in place. This is preferable to dropping the view, losing all the security settings and properties, and then recreating the view. The ALTER command supplies a new SQL SELECT statement for the view: ALTER SchemaName.ViewName AS SQL Select Statement; Management Studio can automatically generate an ALTER statement from an existing view. In Object Explorer, select the view and then choose Script View as ➪ Alter to ➪ New Query Editor Window from the context menu. If the view is no longer needed, it can be completely erased from the database using the DROP command: DROP VIEW SchemaName.ViewName; Within a script that is intended to be executed several times, the following code can drop and recreate the view: IF OBJECT_ID(’vEmployeeList’) IS NOT NULL DROP VIEW dbo.vEmployeeList Go CREATE VIEW SchemaName.ViewName AS SQL Select Statement; Just to reiterate, views don’t contain any data, so there’s no danger that dropping a view will cause any data loss. However, applications, reports, and other objects might depend on the view, and dropping the view might break something else. For more about viewing dependencies within SQL Server, see the section ‘‘Nesting Views’’ later in this chapter. A Broader Point of View The basic mechanics of creating a view and selecting data from the view are pretty straightforward, but views have their own particular nuances — topics such as sorting data, updating data through a view, and nesting views several levels deep. This section examines views from a broader point of view. 336 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 337 Projecting Data Through Views 14 Column aliases The column aliases option is rarely used. With syntax similar to the column list for a common table expression, the view’s column list renames every output column just as if every column had those alias names in the SELECT statement. The view’s column list names override any column names or column aliases in the view’s SELECT statement. The following query alters the vEmployeeList view so that the result columns become ID, Last, First,andJob: ALTER VIEW dbo.vEmployeeList (ID, Last, First, Job) AS SELECT P.BusinessEntityID, P.LastName, P.FirstName, E.JobTitle FROM Person.Person P INNER JOIN HumanResources.Employee E ON P.BusinessEntityID = E.BusinessEntityID GO SELECT * FROM dbo.vEmployeeList Result (abbreviated): ID Last First Job 1S ´ anchez Ken Chief Executive Officer 2 Duffy Terri Vice President of Engineering 3 Tamburello Roberto Engineering Manager 4 Walters Rob Senior Tool Designer Order by and views Views serve as data sources for other queries and don’t support sorting the data within the view. To sort data from a view, include the ORDER BY clause in the query referencing the view. For example, the fol- lowing code selects data from the vEmployeeList view and orders it by LastName, FirstName.The ORDER BY clause is not a part of vEmployeeList , but it is applied to the view by the executing SQL statement: SELECT * FROM dbo.vEmployeeList ORDER BY LastName, FirstName Result: BusinessEntityID LastName FirstName JobTitle 285 Abbas Syed Pacific Sales Manager 38 Abercrombie Kim Production Technician - WC60 211 Abolrous Hazem Quality Assurance Manager 121 Ackerman Pilar Shipping and Receiving Supervisor 337 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 338 Part II Manipulating Data with Select If the view includes a TOP predicate, then the view is allowed to include an ORDER BY — without the ORDER BY,thetop would be meaningless. However, this ORDER BY clause serves only to define which rows qualify for the TOP predicate. The only way to logically guarantee sorted results is to define the ORDER BY clause in the executing query. SQL Server 2000, and some service packs of SQL Server 2005, had a bug (yes, I call it a bug) in the Query Optimizer that would allow an ORDER BY in a view using a top 100 percent predicate. This behavior was never documented or officially supported. However, in SQL Server 2008, this error was corrected and the top 100 percent with an ORDER BY trick will not sort the result. A source of confusion is that Management Studio’s Query Designer allows views to have sorted columns, and it adds the top 100 percent trick to the view. That is a SQL Server 2008 bug. View restrictions Although a view can contain nearly any valid SELECT statement, a few basic restrictions do apply: ■ Views may not include the SELECT INTO option that creates a new table from the selected columns. SELECT INTO fails if the table already exists and it does not return any data, so it’s not a valid view: SELECT * INTO Table ■ Views may not refer to a temporary table (one with a # in the name) or a table variable (preceded with an @), because these types of tables are very transient. ■ The OPTION clause, which gives table or query hints for the entire query, is not allowed. ■ The tablesample table option, which can randomly select pages, is not allowed within aview. ■ Views may not contain compute or compute by columns. Instead, use standard aggregate functions and groupings. ( Compute and compute by are obsolete and are included for backward compatibility only.) Nesting views Because a view is nothing more than a SQL SELECT statement, and a SQL SELECT statement may ref- erence any data source, views may reference other views. Views referred to by other views are sometimes called nested views. The following view uses vEmployeeList and adds a WHERE clause to restrict the results to the smartest and best-looking employees: CREATE VIEW dbo.vEmployeeListDBA AS SELECT BusinessEntityID, LastName, FirstName, JobTitle FROM dbo.vEmployeeList AS vE WHERE JobTitle = ‘Database Administrator’; 338 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 339 Projecting Data Through Views 14 In this example, the view vEmployeeList is nested within vEmployeeListDBA.Anotherwayto express the relationship is to say that vEmployeeListDBA depends on vEmployeeList. Dependencies from other objects in SQL Server can be easily viewed using Object Explorer’s view con- text menu ➪ View Dependencies. Figure 14-3 shows the Object Dependencies dialog for a nested view. FIGURE 14-3 The dependency chain for nested views is easily seen in the Object Dependencies dialog. Here, the vEmployeeListDBA includes the nested view vEmployeeList, which in turn is based on the Employee table, and so on. From code dependencies may be seen using the sys.dm_sql_referencing_entities()function. For example, the following query would indicate whether any other SQL Server object referenced vEmployeeList: SELECT * FROM sys.dm_sql_referencing_entities (’dbo.vEmployeeList, ‘Object’) 339 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 340 Part II Manipulating Data with Select Best Practice W hile there may be a good reason for nesting views to support power users who build ad hoc queries, I don’t recommend nesting views as a general practice. They’re just too difficult to diagnose and maintain. I’ve seen development shops that build their production abstraction layer with nested views several layers deep. It performs poorly and is very difficult to modify. For other options to nesting subselects within outer queries, see Chapter 11, ‘‘Including Data with Subqueries and CTEs.’’ Updating through views One of the main complaints concerning views is that unless the view is a simple single table view, it’s difficult to update the underlying data through the view. While the SQL Server Query Optimizer can update through some complex views, there are some hard-and-fast limitations. Best Practice I don’t recommend designing an application around updatable views. Views are best used as an abstraction layer for ad hoc queries and reports, not for power users to update data, and certainly not for forms, websites, or client applications to update the database. Any of the following factors may cause a view to be non-updatable: ■ Only one table may be updated. If the view includes joins, then the UPDATE statement that references the view must change columns in only one table. ■ Aggregate functions or GROUP BYs in the view will cause the view to be non-updatable. SQL Server couldn’t possibly determine which of the summarized rows should be updated. ■ If the view includes a subquery as a derived table, and any columns from the subquery are exposed as output from the view, then the view is not updateable. However, aggregates are permitted in a subquery that is being used as a derived table, so long as any columns from the aggregate subquery are not in the output columns of the view. ■ If the view includes the WITH CHECK OPTION,theINSERT or UPDATE operation must meet the view’s WHERE-clause conditions. Of course, the other standard potential difficulties with updating and inserting data still apply. Chapter 16, ‘‘Modification Obstacles,’’ discusses in more detail potential troubles with modifying data. One way to work around non-updatable views is to build an INSTEAD OF trigger that inspects the modified data and then performs a legal UPDATE operation based on that data. Chapter 26, ‘‘Creating DML Triggers,’’ explains how to create an INSTEAD OF trigger. 340 www.getcoolebook.com Nielsen c14.tex V4 - 07/21/2009 12:49pm Page 341 Projecting Data Through Views 14 Views and performance Views have an undeserved reputation for poor performance. I think the reason for this belief is based on several factors: ■ Views are often used by power users who submit ad hoc SQL. In earlier versions of SQL Server, ad hoc SQL didn’t perform as well as stored procedures. ■ Views are often used by power users who use front-end UI applications to select and browse data. Some of these applications opened the connections and held locks, causing all sorts of performance problems. ■ Views are often used by power users who find useful data in a view and then build new views on top of views. These nested views might contain a horribly complex view several layers deep that kills performance, while the top-level view appears to be a simple, easy view. Let me put the myth to rest: Well-written views will perform well. The reason to limit views to ad hoc queries and reports isn’t for performance, but for extensibility and control. Alternatives to Views B esides views, SQL Server offers several technologies to build an abstraction layer around the data. Stored procedures are generally my first choice when exposing any data to the outside world. User-defined functions offer several benefits, and inline table-valued user-defined functions are very similar to views but with parameters. Chapter 21 discuss T-SQL, stored procedures, and functions. If you are using views to support ad hoc queries, as I suggest you do, you may also want to explore providing Analysis Services cubes for those users who need to perform complex explorations of the data. Cubes pre-aggregate , or summarize, the data along multiple dimensions. The user may then browse the cube and compare the different data dimensions. For the developer, providing one cube can often eliminate several queries or reports. Chapter 71, ‘‘Building Multidimensional Cubes with Analysis Services,’’ explains how to create cubes. Locking Down the View Views are designed to control access to data. There are several options that protect the data or the view. The WITH CHECK OPTION causes the WHERE clause of the view to check the data being inserted or updated through the view in addition to the data being retrieved. In a sense, it makes the WHERE clause a two-way restriction. The WITH CHECK OPTION is useful when the view should limit inserts and updates with the same restrictions applied to the WHERE clause. 341 www.getcoolebook.com . statement. Any SQL SELECT statement, as long as it’s a valid SQL SELECT state- ment (with a few minor exceptions), can be cut and pasted from nearly any other tool into a view. Within SQL Server Management. placeholder for the subquery. The SQL for the subquery is then manually entered in the SQL pane. The Verify SQL Syntax button in the toolbar verifies only the SQL syntax; it does not verify the. guarantee sorted results is to define the ORDER BY clause in the executing query. SQL Server 2000, and some service packs of SQL Server 2005, had a bug (yes, I call it a bug) in the Query Optimizer that