An SQL text by Erland Sommarskog, SQL Server MVP. Latest revision: 2015-04-14. Copyright applies to this text.
An earlier version of this article is also available in German. Translations provided by SQL Server MVP Frank Kalis.
If you follow the various newsgroups on Microsoft SQL Server, you often see people asking why they can't do:
SELECT * FROM @tablename SELECT @colname FROM tbl SELECT * FROM tbl WHERE x IN (@list)
For all three examples you can expect someone to answer Use dynamic SQL and give a quick example on how to do it. Unfortunately, for all three examples above, dynamic SQL is a poor solution. On the other hand, there are situations where dynamic SQL is the best or only way to go.
In this article I will discuss the use of dynamic SQL in stored procedures and to a minor extent from client languages. To set the scene, I start with a very quick overview on application architecture for data access. I then proceed to describe the feature dynamic SQL as such, with a quick introduction followed by the gory syntax details. Next, I continue with a discussion on SQL injection, a security issue that it is essential to have good understanding of when you work with dynamic SQL. This is followed by a section where I discuss why we use stored procedures, and how that is affected by the use of dynamic SQL. I carry on with a section on good practices and tips for writing dynamic SQL. I conclude by reviewing a number of situations where you could use dynamic SQL and whether it is a good or bad idea to do it.
The article covers all versions of SQL Server from SQL 6.5 to SQL 2008, with emphasis on SQL 2000 and later versions.
Note: many of
the code samples in this text works against the pubs and Northwind databases
that ship with SQL 2000 and SQL 7, but not with SQL 2005
and later. You can download
these databases from
Microsoft's web site.
Before I describe dynamic SQL, I like to briefly discuss the various ways you can access data from an application to give an overview of what I'll be talking about in this article.
(Note: all through this text I will refer to client as anything that accesses SQL Server from the outside. In the overall application architecture that may in fact be a middle tier or a business layer, but as that is of little interest to this article, I use client in the sake of brevity.)
There are two main roads to go, and then there are forks and sub-forks.
Fork 1-a may be good for simple tasks, but you are likely to find that you outgrow it as the complexity of your application increases. In any case, this approach falls entirely outside the scope of this article.
Many applications are built along the principles of fork 1-b, and as long as you take the sub-fork 1-b-ii, it does not have to be bad. (Why 1-b-i is bad, is something I will come back to. Here I will just drop two keywords: SQL Injection and Query-Plan Reuse.) Nonetheless, in many shops the mandate is that you should use stored procedures. When you use stored procedures with only static SQL, users do not need direct permissions to access the tables, only permissions to execute the stored procedures, and thus you can use the stored procedure to control what users may and may not do.
The main focus for this text is sub-fork 2-a-ii. When used appropriately, dynamic SQL in stored procedures can be a powerful addition to static SQL. But some of the questions on the newsgroups leads to dynamic SQL in stored procedures that are so meaningless, that these people would be better off with fork 1-b instead.
Finally, fork 2-b, stored procedures in the CLR, is in many regards very similar to fork 1-b, since all data access from CLR procedures is through generated SQL strings, parameterised or unparameterised. If you have settled on SQL procedures for your application, there is little point in rewriting them into the CLR. However, CLR code can be a valuable supplement for tasks that are difficult to perform in T-SQL, but you yet want to perform server-side.
In this chapter I will first look at some quick examples of dynamic SQL and point out some very important implications of using dynamic SQL. I will then describe sp_executesql and EXEC() in detail, the two commands you can use to invoke dynamic SQL from T-SQL.
Understanding dynamic SQL itself is not difficult. Au contraire, it's rather too easy to use. Understanding the fine details, though, takes a little longer time. If you start out using dynamic SQL casually, you are bound to face accidents when things do not work as you have anticipated.
One of the problems listed in the introduction was how to write a stored procedure that takes a table name as its input. Here are two examples, based on the two ways to do dynamic SQL in Transact-SQL:
CREATE PROCEDURE general_select1 @tblname sysname, @key varchar(10) AS DECLARE @sql nvarchar(4000) SELECT @sql = ' SELECT col1, col2, col3 ' + ' FROM dbo.' + quotename(@tblname) + ' WHERE keycol = @key' EXEC sp_executesql @sql, N'@key varchar(10)', @key
CREATE PROCEDURE general_select2 @tblname nvarchar(127), @key varchar(10) AS EXEC('SELECT col1, col2, col3 FROM ' + @tblname + ' WHERE keycol = ''' + @key + '''')
Before I say anything else, permit me to point out that these are examples of bad usage of dynamic SQL. Passing a table name as a parameter is not how you should write stored procedures, and one aim of this article is to explain this in detail. Also, the two examples are not equivalent. While both examples are bad, the second example has several problems that the first does not have. What these problems are will be apparent as you read this text.
Whereas the above looks very simple and easy, there are some very important things to observe. The first thing is permissions. You may know that when you use stored procedures, users do not need permissions to access the tables accessed by the stored procedure. This does not apply when you use dynamic SQL! For the procedures above to execute successfully, the users must have SELECT permission on the table in @tblname. In SQL 2000 and earlier this is an absolute rule with no way around it. Starting with SQL 2005, there are alternatives, something I will come back to in the section The Permission System.
Next thing to observe is that the dynamic SQL is not part of the stored procedure, but constitutes its own scope. Invoking a block of dynamic SQL is akin to call a nameless stored procedure created ad-hoc. This has a number of consequences:
As you've seen there are two ways to invoke dynamic SQL, sp_executesql and EXEC(). sp_executesql was added in SQL 7, whereas EXEC() has been around since SQL 6.0. In application code, sp_executesql should be your choice 95% of the time for reasons that will prevail. For now I will only give two keywords: SQL Injection and Query-Plan Reuse. EXEC() is mainly useful for quick throw-away things and DBA tasks, but also comes to the rescue in SQL 2000 and SQL 7 when the SQL string exceeds 4000 characters. And, obviously, in SQL 6.5, EXEC() is the sole choice. In the next two sections we will look at these two commands in detail.
sp_executesql is a built-in stored procedure that takes two pre-defined parameters and any number of user-defined parameters.
The first parameter @stmt is mandatory, and contains a batch of one or more SQL statements. The data type of @stmt is ntext in SQL 7 and SQL 2000, and nvarchar(MAX) in SQL 2005 and later. Beware that you must pass an nvarchar/ntext value (that is, a Unicode value). A varchar value won't do.
The second parameter @params is optional, but you will use it 90% of the time. @params declares the parameters that you refer to in @stmt. The syntax of @params is exactly the same as for the parameter list of a stored procedure. The parameters can have default values and they can have the OUTPUT marker. Not all parameters you declare must actually appear in the SQL string. (Whereas all variables that appear in the SQL string must be declared, either with a DECLARE inside @stmt, or in @params.) Just like @stmt, the data type of @params is ntext SQL 2000 and earlier and nvarchar(MAX) since SQL 2005.
The rest of the parameters are simply the parameters that you declared in @params, and you pass them as you pass parameters to a stored procedure, either positional or named. To get a value back from your output parameter, you must specify OUTPUT with the parameter, just like when you call a stored procedure. Note that the first two parameters, @stmt and @params, must be specified positionally. You can provide the parameter names for them, but these names are blissfully ignored.
Let's look at an example. Say that in your database, many tables have a column LastUpdated, which holds the time a row last was updated. You want to be able to find out how many rows in each table that were modified at least once during a period. This is not something you run as part of the application, but something you run as a DBA from time to time, so you just keep it as a script that you have a around. Here is what it could look like:
DECLARE @tbl sysname, @sql nvarchar(4000), @params nvarchar(4000), @count int DECLARE tblcur CURSOR STATIC LOCAL FOR SELECT object_name(id) FROM syscolumns WHERE name = 'LastUpdated' ORDER BY 1 OPEN tblcur WHILE 1 = 1 BEGIN FETCH tblcur INTO @tbl IF @@fetch_status <> 0 BREAK SELECT @sql = N' SELECT @cnt = COUNT(*) FROM dbo.' + quotename(@tbl) + N' WHERE LastUpdated BETWEEN @fromdate AND ' + N' coalesce(@todate, ''99991231'')' SELECT @params = N'@fromdate datetime, ' + N'@todate datetime = NULL, ' + N'@cnt int OUTPUT' EXEC sp_executesql @sql, @params, '20060101', @cnt = @count OUTPUT PRINT @tbl + ': ' + convert(varchar(10), @count) + ' modified rows.' END DEALLOCATE tblcur
I've put the lines that pertain directly to the dynamic SQL in bold face. You can see that I have declared the @sql and @params variables to be of the maximum length for nvarchar variables in SQL 2000. In SQL 2005 and later, you may want to make it a routine to declare @sql as nvarchar(MAX), more about this just below.
When I assign the @sql variable, I am careful to format the statement so that
it is easy to read, and I leave in spaces to avoid that two concatenated
parts are glued together without space in between, which could cause a syntax
error. I put the table name in
quotename() in case a table name has any special
characters in it. I also prefix the table name with "dbo.", which is a good habit, as we will see when we look at dynamic SQL and
query plans. Overall, I will cover this sort of
good practices more in detail later in the text. Note also the appearance of
'' around the date literal – the rule in T-SQL is that to include the string
delimiter in a string, you must double it.
In this example, the dynamic SQL has three parameters: one mandatory input parameter, one optional input parameter, and one output parameter. I've assumed that this time the DBA wanted to see all changes made after 2006-01-01, which is why I've left out @todate in the call to sp_executesql. Since I left out one variable, I must specify the last, @cnt by name – the same rules as when you call a stored procedure. Note also that the variable is called @cnt in the dynamic SQL, but @count in the surrounding script. Normally, you might want to use the same name, but I wanted to stress that the @cnt in the dynamic SQL is only visible within the dynamic SQL, whereas @count is not visible there.
You may note that I've prefix the string literals with N to denote that they are Unicode strings. As @sql and @params are declared as nvarchar, technically this is not necessary (as long as you stick to your 8-bit character set). However, when you provide any of the strings directly in the call to sp_executesql, you must specify the N, as in this fairly silly example:
EXEC sp_executesql N'SELECT @x', N'@x int', @x = 2
If you remove any of the Ns, you will get an error message. Since sp_executesql is a built-in stored procedure, there is no implicit conversion from varchar.
You may wonder why I do not pass @tbl as a parameter as well. The answer is that you can't. Dynamic SQL is just like any other SQL. You can't specify a table name through a variable in T-SQL, that's the whole story. Thus, when you need to specify things like table names, column names etc dynamically, you must interpolate them into the string.
If you are on SQL 2000 or SQL 7, there is a limitation with sp_executesql when it comes to the length of the SQL string. While the parameter is ntext, you cannot use this data type for local variables. Thus, you will have to stick to nvarchar(4000). In many cases this will do fine, but it is not uncommon to exceed that limit. In this case, you will need to use EXEC(), described just below.
Since SQL 2005, this is not an issue. Here you can use the new data type nvarchar(MAX) which can hold as much data as ntext, but without the many restrictions of ntext.
EXEC() takes one parameter which is an SQL statement to execute. The parameter can be a concatenation of string variables and string literals, but cannot include calls to functions or other operators. For very simple cases, EXEC() is less hassle than sp_executesql. For instance, say that you want to run UPDATE STATISTICS WITH FULLSCAN on some selected tables. It could look like this:
FETCH tblcur INTO @tbl IF @@fetch_status <> 0 BREAK EXEC('UPDATE STATISTICS [' + @tbl + '] WITH FULLSCAN')
In the example with sp_executesql, I used quotename(), but here I've let it suffice with adding brackets, in case there is a table named Order Details (which there is in the Northwind database). Since EXEC only permits string literals and string variables to be concatenated and not arbitrary expressions, this is not legal:
EXEC('UPDATE STATISTICS ' + quotename(@tbl) + ' WITH FULLSCAN')
Best practice is to always use a variable to hold the SQL statement, so the example would better read:
FETCH tblcur INTO @tbl IF @@fetch_status <> 0 BREAK SELECT @sql = 'UPDATE STATISTICS ' + quotename(@tbl) + ' WITH FULLSCAN' EXEC(@sql)
The fact that you can concatenate strings within EXEC() permits you to make very quick things, which can be convenient at times, but it can lead to poor habits in application code. However, there are situations where this is an enormous blessing. As I mentioned, in SQL 7 and SQL 2000, you can in practice only use 4000 characters in your SQL string with sp_executesql. EXEC does not have this limitation, since you can say:
EXEC(@sql1 + @sql2 + @sql3)
Where all of @sql1, @sql2 and @sql3 can be 4000 characters long – or even 8000 characters as EXEC() permits you to use varchar.
Since you cannot use parameters, you cannot as easily get values out from EXEC() as you can with sp_executesql. You can, however, use INSERT-EXEC to insert the result set from EXEC() into a table. I will show you an example later on, when I also show you how you can use EXEC() to pass longer strings than 4000 characters to sp_executesql.
In SQL 2005 and later, EXEC() permits impersonation so that you can say:
EXEC(@sql) AS USER = 'mitchell' EXEC(@sql) AS LOGIN = 'CORDOBA\Miguel'
This is mainly a syntactical shortcut that saves you from embedding the invocation of dynamic SQL in EXECUTE AS and REVERT. (I discuss these statements more in detail in my article Granting Permissions Through Stored Procedures.)
SQL 2005 adds a valuable extension to EXEC(): you can use it to execute strings on linked servers. I will cover this form of EXEC() in a separate section later in this text.
Before you start to use dynamic SQL all over town, you need to learn about SQL injection and how you protect your application against it. SQL injection is a technique whereby an intruder enters data that causes your application to execute SQL statements you did not intend it to. SQL injection is possible as soon there is dynamic SQL which is handled carelessly, be that SQL statements sent from the client, dynamic SQL generated in T-SQL stored procedures, or SQL batches executed from CLR stored procedures. This is not a line of attack that is unique to MS SQL Server, but all RDBMS are open to it.
Here is an example. The purpose of the procedure below is to permit users to search for orders by various conditions. A real-life example of such a procedure would have many more parameters, but I've cut it down to two to be brief. (This is, by the way, a problem for which dynamic SQL is a very good solution.) As the procedure is written, it is open for SQL injection:
CREATE PROCEDURE search_orders @custid nchar(5) = NULL, @shipname nvarchar(40) = NULL AS DECLARE @sql nvarchar(4000) SELECT @sql = ' SELECT OrderID, OrderDate, CustomerID, ShipName ' + ' FROM dbo.Orders WHERE 1 = 1 ' IF @custid IS NOT NULL SELECT @sql = @sql + ' AND CustomerID LIKE ''' + @custid + '''' IF @shipname IS NOT NULL SELECT @sql = @sql + ' AND ShipName LIKE ''' + @shipname + '''' EXEC(@sql)
Before we look at a real attack, let's just discuss this from the point of view of user-friendliness. Assume that the input for the parameters @custid and @shipname comes directly from the user and a naïve and innocent user wants to look for orders where ShipName is Let's Stop N Shop, so he enters Let's. Do you see what will happen? Because @shipname includes a single quote, he will get a syntax error. So even if you think that SQL injection is no issue to you, because you trust your users, you still need to read this section, so that they can search for Brian O'Brien and Samuel Eto'o.
So this is the starting point. A delimiter, usually a single quote, affects your dynamic SQL, and a malicious user can take benefit of this. For instance, consider this input for @shipname:
' DROP TABLE Orders --
The resulting SQL becomes:
SELECT * FROM dbo.Orders WHERE 1 = 1 AND ShipName LIKE '' DROP TABLE orders --'
This is a perfectly legal batch of T-SQL, including the text in red. Since there is something called permissions in SQL Server, this attack may or may not succeed. A plain user who runs a Windows application and who logs into SQL Server with his own login, is not likely to have permissions to drop a table. But it is not uncommon for web applications to have a general login that runs SQL queries on behalf of the users. And if this web app logs into SQL Server with sysadmin or db_owner privileges, the attack succeeds. Mind you, with sysadmin rights, the attacker can add users and logins as he pleases. And if the service account for SQL Server has admin privileges in Windows, the attacker has access into your network far beyond SQL Server through xp_cmdshell. (Which is disabled by default on SQL 2005 and later, but if the attacker has achieved sysadmin rights on the server, he can change that.)
Typically, an attacker first tests what happens
if he enters a single quote (
') in an input field or a URL. If this
yields a syntax error, the attacker knows that there is a vulnerability. He
then finds out if he needs any extra tokens to terminate the query, and then
he can add his own SQL statement. Finally he adds a comment character to kill
the rest of the SQL string to avoid syntax errors. Single quote is the most
common character to reveal openings for SQL injection, but if you have
dynamic table and column names, there are more options an attacker could
Take this dreadful version of general_select:
CREATE PROCEDURE general_select2 @tblname nvarchar(127), @key varchar(10) AS EXEC('SELECT col1, col2, col3 FROM ' + @tblname + ' WHERE keycol = ''' + @key + '''')
and assume that @tblname comes from a URL. There are quite some options that an attacker could use to take benefit of this hole.
And don't overlook numeric values: they can very well be used for SQL injection. Of course, in a T-SQL procedure where the value is passed as an int parameter there is no risk, but if a supposedly numeric value is directly interpolated into an SQL string in client code, there is a huge potential for SQL injection.
Keep in mind that user input comes from more places than just input fields on a form. The most commonly used area for injection attacks on the Internet is probably parameters in URLs and cookies. Thus, be very careful how you handle anything that comes into your application from the outside.
You may think that it takes not only skill, but also luck for someone to find and exploit a hole for SQL injection. But remember that there are too many hackers out there with too much time on their hands. SQL injection is a serious security issue, and you must take precautions to protect your applications against it.
One approach I seen mentioned from time to time, is to validate input data in some way, but in my opinion that is not the right way to go. Here are are the three steadfast principles you need to follow:
The first point is mainly a safeguard, so that if there is a injection hole, the intruder will not be able to do that much harm. The second point makes the task for the attacker more difficult as he cannot get feedback from his attempts.
But it is the third point that is the actual protection, and that we will look a little closer at. The procedure search_orders above should be coded as:
CREATE PROCEDURE search_orders @custid nchar(5) = NULL, @shipname nvarchar(40) = NULL AS DECLARE @sql nvarchar(4000) SELECT @sql = ' SELECT OrderID, OrderDate, CustomerID, ShipName ' + ' FROM dbo.Orders WHERE 1 = 1 ' IF @custid IS NOT NULL SELECT @sql = @sql + ' AND CustomerID LIKE @custid ' IF @shipname IS NOT NULL SELECT @sql = @sql + ' AND ShipName LIKE @shipname ' EXEC sp_executesql @sql, N'@custid nchar(5), @shipname nvarchar(40)', @custid, @shipname
Since the SQL string does not include any user input, there is no opening for SQL injection. It's as simple as that. By the way, note that since we can include parameters in the parameter list, even if they don't actually appear in the SQL string, we don't need any complicated logic to build the parameter list, but can keep it static. In the same vein, we can always pass all input parameters to the SQL string.
As you may recall, you cannot pass everything as parameters to dynamic SQL, for instance table and column names. In this case you must enclose all such object names in quotename(), that I will return to in the section Good Coding Practices and Tips for Dynamic SQL.
The example above was for dynamic SQL in a T-SQL stored procedure. The same advice applies to SQL generated in client code or in a CLR stored procedure. Since this is so important, here is an example of coding the above in VB6 and ADO:
Set cmd = CreateObject("ADODB.Command") Set cmd.ActiveConnection = cnn cmd.CommandType = adCmdText cmd.CommandText = " SELECT OrderID, OrderDate, CustomerID, ShipName " & _ " FROM dbo.Orders WHERE 1 = 1 " If custid <> "" Then cmd.CommandText = cmd.CommandText & " AND CustomerID LIKE ? " cmd.Parameters.Append cmd.CreateParameter("@custid", adWChar, adParamInput, 5, custid) End If If shipname <> "" Then cmd.CommandText = cmd.CommandText & " AND ShipName LIKE ? " cmd.Parameters.Append cmd.CreateParameter("@shipname", _ adVarWChar, adParamInput, 40, shipname) End If Set rs = cmd.Execute
Since the main focus of this text is dynamic SQL in T-SQL procedures, I will
explain this example only briefly. In ADO you use
? as a parameter
marker, and you can only pass parameters that
actually appear in the SQL string. (If you
specify too many parameters, you will get a completely incomprehensible error
message.) If you use the SQL Profiler to see what ADO
sends to SQL Server, you will find that it invokes – sp_executesql.
Protection against SQL injection is not the only advantage of using parameterised queries. In the section Caching Query Plans, we will look more in detail on parameterised queries and at a second very important reason to use them. This section also includes an example of composing and sending a parameterised SQL statement for SqlClient in VB .Net.
You may think that an even better protection against SQL injection is to use stored procedures with static SQL only. Yes, this is true, but! It depends on how you call your stored procedures from the client. If you compose an EXEC command into which you interpolate the input values, you are back on square one and you are as open to SQL injection as ever. In ADO, you need to call your procedure with the command type adCmdStoredProc and use .CreateParameter to specify the parameters. By specifying adCmdStoredProc, you call the stored procedure through RPC, Remote Procedure Call, which not only protects you against SQL injection, but it is also more efficient. Similar measures apply to other client APIs; all APIs I know of supply a way to call a stored procedure through RPC.
In the introduction, I presented various strategies for data-access for an application, and I said that in many shops all data access is through stored procedures. In this section, I will look a little closer at the advantages with using stored procedures over sending SQL statements from the client. I will also look at what happens when you use dynamic SQL in a stored procedure, and show that you lose some of the advantages with stored procedures, whereas other are unaffected.
Historically, using stored procedures has been the way to give users access to data. In a locked-down database, users do not have permissions to access tables directly. Instead, the application performs all access through stored procedures that retrieve and update data in a controlled way, so that users only get to see data they have access to, and they cannot perform updates that violate business rules. This works as long as the procedure and the tables have the same owner, typically dbo (the database owner), through a mechanism known as ownership chaining.
As I have already mentioned, ownership chaining does not work when you use dynamic SQL. The reason for this is very simple: the block of dynamic SQL is not a procedure and does not have any owner. Thus the chain is broken.
In SQL 2005 and later versions of SQL Server, this can be addressed by signing a procedure that uses dynamic SQL with a certificate. You associate the certificate with a user, and grant that user (which is a user that cannot log in) the rights needed for the dynamic SQL to execute successfully. A second method is to use the EXECUTE AS clause to impersonate a user that has been granted the necessary permissions. This method is easier to use, but has side effects that can have unacceptable consequences for auditing, row-level security schemes and system monitoring. For this reason, my strong recommendation is to use certificates.
Describing these methods more closely, would take up too much space here. Instead I've written a separate article about them, Giving Permissions through Stored Procedures, where I discusses both certificates and impersonation in detail, and I also take a closer look on ownership chaining.
If you write CLR procedures that perform data access, the same is true for them. Ownership chaining never applies since all data access in a CLR procedure is through dynamic SQL. But you can use certificates or impersonation to avoid having to give users direct permissions on the tables.
On SQL 2000 there is no way to combine dynamic SQL with the encapsulation of permissions that you can get through stored procedures. Any use of dynamic SQL requires that the users have direct permissions on the accessed tables. If your security scheme precludes giving users permissions to access tables directly, you cannot use dynamic SQL. It is that plain and simple. Depending on the sensitivity of the data in the application, it may be acceptable to give the users SELECT permissions on the tables (or on some tables) to permit the use of dynamic SQL. I strongly recommend against granting users INSERT, UPDATE and DELETE rights on tables only to permit dynamic SQL in some occasional procedure.
There are however, some ways to arrange so that users only have access to the data through the application. All and all, there are three alternatives, application roles, "application proxies" and Terminal Server. All require you to change the application architecture or infrastructure, so it is nothing you introduce at whim.
Application roles were introduced in SQL 7. Users log into SQL Server but have no permissions on their own beyond the database access. Instead, the application activates the application role by sending a password somehow embedded into it, and this application role has the permissions needed. With "application proxies", the application authenticates the users outside SQL Server and logs into SQL Server on their behalf with a proxy login. This proxy login impersonates the users in SQL Server, and thus their permissions apply. However, since the users do not have any login on their own, they cannot log into SQL Server outside the application. In Giving Permissions..., I discuss these two methods a little further.
The final possibility is to put the application on Terminal Server. Users log into the terminal server which is set up so that all they can do is to run this application. Furthermore, the network is configured so that they cannot access SQL Server from their regular computers. Thus, the application is their only way to the data.
For all these methods, keep in mind about SQL injection, and do not grant more permissions than needed.
Every query you run in SQL Server requires a query plan. When you run a query the first time, SQL Server builds a query plan for it – or as the terminology goes – it compiles the query. SQL Server saves the plan in cache, and next time you run the query, the plan is reused. The query plan stays in cache until it's aged out because it has not been used for a while, or it is invalidated for some reason. (Why this happens falls outside the scope of this article.)
The reuse of cached query plans is very important for the performance of queries where the compilation time is in par with the execution time or exceeds it. If a query needs to run for four minutes, it does not matter much if the query is recompiled for an extra second each time. On the other hand, if the execution time of the query is 40 ms but it takes one second to compile the query, there is a huge gain with the cached plan, particularly if the query is executed over and over again.
Up to SQL 6.5 the only plans there were put into the cache were plans for stored procedures. Loose batches of SQL were compiled each time. And since the query plan for dynamic SQL is not part of the stored procedure, that included dynamic SQL as well. Thus in SQL 6.5, the use of dynamic SQL nullified the benefit with stored procedures in this regard.
Starting with SQL 7, SQL Server also caches the plans for bare statements sent from a client or generated through dynamic SQL. Say that you send this query from the client, or execute it with EXEC():
SELECT O.OrderID, SUM(OD.UnitPrice * OD.Quantity) FROM Orders O JOIN [Order Details] OD ON O.OrderID = OD.OrderID WHERE O.OrderDate BETWEEN '19980201' AND '19980228' AND EXISTS (SELECT * FROM [Order Details] OD2 WHERE O.OrderID = OD2.OrderID AND OD2.ProductID = 76) GROUP BY O.OrderID
The query returns the total order amount for the orders in February 1998 that contained the product Lakkalikööri. SQL Server will put the plan into the cache, and next time you run this query, the plan will be reused. But only if it is exactly the same query. Since the cache lookup is by a hash value computed from the query text, the cache is space- and case-sensitive. Thus, if you add a single space somewhere, the plan is not reused. More importantly, it is not unlikely that next time you want to run the query for a different product, or a different period.
All this changes, if you instead use sp_executesql to run your query with parameters:
DECLARE @sql nvarchar(2000) SELECT @sql = 'SELECT O.OrderID, SUM(OD.UnitPrice * OD.Quantity) FROM dbo.Orders O JOIN dbo.[Order Details] OD ON O.OrderID = OD.OrderID WHERE O.OrderDate BETWEEN @from AND @to AND EXISTS (SELECT * FROM dbo.[Order Details] OD2 WHERE O.OrderID = OD2.OrderID AND OD2.ProductID = @prodid) GROUP BY O.OrderID' EXEC sp_executesql @sql, N'@from datetime, @to datetime, @prodid int', '19980201', '19980228', 76
The principle for cache lookup is the same as for a non-parameterised query: SQL Server hashes the query text and looks up the hash value in the cache, still in a case- and space-sensitive fashion. But since the parameter values are not part of the query text, the same plan can be reused even when the input changes.
To make this really efficient there is one more thing you need to observe. Do you see that I've prefixed all tables in the query with dbo? There is a very important reason for this. Users can have different default schema, and up to SQL 2000, all users had a default schema equal to their username. Thus, if default schema for user1 is user1, and this users runs a query that goes "SELECT ... FROM Orders", SQL Server must first check if there is a table user1.Orders, before it looks for dbo.Orders. Since user1.Orders could appear on the scene at any time, user1 cannot share cache entry with a user different default schema. Yes, in SQL 2005, it is perfectly possible that all users have dbo as their default schema, but it seems to be a bad idea to rely on it.
If you instead use stored procedures, it is not equally important to prefix tables with dbo. Microsoft still recommends that you do, but even if you don't, users with different default schema can share the same query plan.
From what I have said here, it follows that if you use dynamic SQL with EXEC() you lose an important benefit of stored procedures whereas with sp_executesql you don't. At least in theory. It's easy to forget that dbo, and if you leave it out in just a single place in the query, you will get as many entries in the cache for the query as there are users running it. Recall also that the cache is space- and case-sensitive, so if you generate the same query in several places, you may inadvertently have different spacing or inconsistent use of case. And this is not restricted to the SQL statement, the parameter list is as much part of the cache entry. Furthermore, since the cache lookup is by a hash value computed from the query text, I would assume that this is somewhat more expensive than looking up a stored procedure. In fact, under extreme circumstances, heavy use of dynamic SQL, can lead to serious performance degradation. Some of my MVP colleagues have observed systems with lots of memory (> 20 GB) when the plan cache has been so filled with plans for SQL statements, that there have been hash collisions galore, and the cache lookup alone could take several seconds. Presumably, the applications in question either did not use parameterised queries at all, or they failed to prefix tables with dbo.
So far, I've only talked about dynamic SQL in stored procedures. But in this regard there is very little difference to SQL statements sent from the client, or SQL statements generated in CLR procedures. The same rules apply: unparameterised statements are cached but with little probability for reuse, whereas parameterised queries can be as efficient as stored procedures if you remember to always prefix the tables with dbo. (And still with the caveat that the cache lookup is space- and case-sensitive.) Most client APIs implement parameterised queries by calling sp_executesql under the covers.
In the section on SQL Injection, I included an example on how to do parameterised queries with ADO and VB6. Here is an example with VB .Net and SqlClient:
cmd.CommandType = System.Data.CommandType.Text cmd.CommandText = _ " SELECT O.OrderID, SUM(OD.UnitPrice * OD.Quantity)" & _ " FROM dbo.Orders O " & _ " JOIN dbo.[Order Details] OD ON O.OrderID = OD.OrderID" & _ " WHERE O.OrderDate BETWEEN @from AND @to" & _ " AND EXISTS (SELECT *" & _ " FROM dbo.[Order Details] OD2" & _ " WHERE O.OrderID = OD2.OrderID" & _ " AND OD2.ProductID = @prodid)" & _ " GROUP BY O.OrderID" cmd.Parameters.Add("@from", SqlDbType.Datetime) cmd.Parameters("@from").Value = "1998-02-01" cmd.Parameters.Add("@to", SqlDbType.Datetime) cmd.Parameters("@to").Value = "1998-02-28" cmd.Parameters.Add("@prodid", SqlDbType.Int) cmd.Parameters("@prodid").Value = 76
In contrast to ADO, SqlClient uses names with @ for parameters. The syntax for defining parameters is similar to ADO, but not identical. This article is long enough, so I will not go into details on how the Parameters collection works. Instead, I refer you to MSDN where both SqlClient and ADO are documented in detail. Whatever client API you are using, please learn how to use parameterised commands with it. Yes, there is a tone of desperation in my voice. I don't know how many posts I've seen on the newsgroups over the years where people build their SQL strings by interpolating the values from input fields into the SQL string, and thereby degrading the performance of their application, and worst of all opening their database to SQL injection.
... and just when you thought you were safe, I need to turn this upside down. Recall what I said in the beginning of this section, that if the query is going to run for four minutes, one second extra for compilation is not a big deal. And if that recompilation slashes the execution time from forty minutes to four, there is a huge gain. Most queries benefit from cached parameterised plans, but not all do. Say that you have a query where the user can ask for data for some time span. If the user asks for a summary for a single day, there is a good non-clustered index that can be used for a sub-second response time. But if the request is for the entire year, the same index would be a disaster, and a table scan is better. Starting with SQL 2005 you can force a query to be recompiled each time it is executed by adding OPTION (RECOMPILE) to the end of the query, and thus you can still use sp_executesql to get the best protection against SQL injection. On SQL 2000 and earlier, it may in fact be better to interpolate critical parameters into the query string when you need to force recompilation each time.
For the sake of completeness, I should mention that SQL Server is able to auto-parameterise queries. If you submit:
SELECT OrderID, OrderDate FROM dbo.Orders WHERE CustomerID = N'ALFKI'
SQL Server may recast this as
SELECT OrderID, OrderDate FROM dbo.Orders WHERE CustomerID = @P1
so if next time you submit BERGS instead of ALFKI, the query plan will be reused. Auto-parameterisation comes in two flavours: simple and forced. Simple is the default and is the only option on SQL 2000 and earlier. With simple parameterisation, auto-parameterisation happens only with very simple queries, and, it seems, with some inconsistency. With forced parameterisation, SQL Server parameterises all queries that comes its way (with some exceptions documented in Books Online). Forced parameterisation is, in my opinion, mainly a setting to cover up for poorly designed third-party application that uses unparameterised dynamic SQL. For your own development you should not rely on any form of auto-parameterisation. (But in the situation you really a want a new query plan each time, you may have to verify that it doesn't happen when you don't want to.)
They say seeing is believing. Here is a demo that you can try on yourself, if you have SQL 2005. First create this database:
CREATE DATABASE many_sps go USE many_sps go DECLARE @sql nvarchar(4000), @x int SELECT @x = 200 WHILE @x > 0 BEGIN SELECT @sql = 'CREATE PROCEDURE abc_' + ltrim(str(@x)) + '_sp @orderid int AS SELECT O.OrderID, O.OrderDate, O.CustomerID, C.CompanyName, Prodcnt = OD.cnt, Totalsum = OD.total FROM Northwind..Orders O JOIN Northwind..Customers C ON O.CustomerID = C.CustomerID JOIN (SELECT OrderID, cnt = COUNT(*), total = SUM(Quantity * UnitPrice) FROM Northwind..[Order Details] GROUP BY OrderID) AS OD ON OD.OrderID = O.OrderID WHERE O.OrderID = @orderid' EXEC(@sql) SELECT @x = @x - 1 END
Then in SQL Server Management Studio 2005, press F7 navigate down to the list of stored procedures. Select all procedures. Then from the context menu select to script them as CREATE TO to a new query window. How long time this takes depends on your hardware, but on my machine it took 90 seconds and at the same time SQL Server grabbed over 250 MB of memory. If you use the Profiler to see what Mgmt Studio is up to, you will see that for each procedure, Mgmt Studio emits a couple of queries with the procedure name embedded. That is, no parameterised statements. Once scripting is complete, issue this command:
ALTER DATABASE many_sps SET PARAMETERIZATION FORCED
and redo the operation. On my machine scripting now completed in five seconds!. This demonstrates that the difference between parameterised and unparameterised can be dramatic. (And that Microsoft can not use their own products properly.) If you run SQL Server on your local machine, you can see this from one more angle, you can stop and restart SQL Server before the two scripting operations, and then use Task Manager to see how much physical memory SQL Server uses in the two cases. That difference lies entirely in the plan cache.
This particular issue have been addressed in SQL Server Management Studio 2008. SSMS 2008 has its own scripting issues, but they have nothing to do with the topic of this article.
Another advantage with stored procedures over SQL sent from the client is that less bytes travel the network. Rather than sending a 50-line query over the network, you only need to pass the name of a stored procedure and a few parameters. This gets more significant if the computation requires several queries, possibly with logic in between. If all logic is outside the database, this could mean that data has to travel up to the client, only to travel back in the next moment. With stored procedures you can use temp tables to hold intermediate results. (You can use temp tables from outer layers as well, although it may require some careful use of your client API.)
In this case, the dividing line goes between sending SQL from the client or running stored procedures. If the stored procedures use static SQL only, or invoke dynamic SQL does not matter, nor does it matter if it is a CLR procedure. You still get the gains of reduced network traffic.
This is not a question of security or performance, but one of good programming practice and modularising your code. By using stored procedures, you don't have to bog down your client code with the construction of SQL statements. Then again, it depends a little on what you put into those stored procedure. Myself, I am of the school that the business logic should be where the data is, and in this case there is no dispute that you should use stored procedures to encapsulate your logic.
But there are also people who like to see the database as a unintelligent container of data, and who prefer to have the business logic elsewhere. In this case, the arguments for using stored procedures for encapsulation may not be equally compelling. You could just as well employ careful programming practices in your client language and send SQL strings.
Nothing of this changes if you use dynamic SQL in your stored procedures. The stored procedure is still a container for some piece of logic, and how it looks on the inside does not matter. I'm here assuming that most of your procedures use static SQL only. If all your stored procedures generate dynamic SQL, then you are probably better off in this regard to do it all in client code. Then again, sometimes there is no other application than Query Analyzer or SQL Server Management Studio. (Typically this would be tasks that are run by an admin.) In this case, the only container of logic available is stored procedures, and it's immaterial whether they use dynamic SQL or not.
In a complex system with hundreds of tables, you may need to know where a certain table or column is referenced, because you are considering changing or dropping it. If all access to tables is from static SQL in stored procedures, you may be able find all references by using the system stored procedure sp_depends or query a system table directly. (sysdepends in SQL 2000, sys.sql_dependencies in SQL 2005 and later. In SQL 2008 there is also sys.sql_expression_dependencies.) I say may, because it is very difficult to maintain complete dependency information in SQL Server. If you drop and recreate a table, all dependency information for the table is lost. What I do myself is to regularly build an empty database from our version-control system, and since our build tool loads all tables before any stored procedure or trigger, I know that I can trust the dependency information in that database.
If you throw dynamic SQL into the mix – be that SQL sent from client, dynamic SQL in T-SQL procedures, or SQL generated by CLR stored procedures - you lose this opportunity. The alternative is to employ brute-force search, and if the construction of dynamic SQL is confined to some well-defined set of modules, this may work. If not, you may end up with a database where no one ever dares to drop or change a column or a table, and which eventually becomes unbearable complex and inefficient because of all the legacy baggage it's carrying around.
While the main dividing line here is between static SQL and any form of dynamic SQL, dynamic SQL in T-SQL stored procedures is probably the least harmful, as there is less code to search. You can even search the column sys.sql_modules.definition using SQL. Available since SQL 2005. In SQL 2000 you can search syscomments, but as the procedure text there is chopped into 4000-char slices, this is less reliable.
In any case, an occasional stored procedure that uses dynamic SQL is not likely cause the Armageddon I pictured above. But it is a good argument for being restrictive with dynamic SQL in any form.
One distinct advantage of writing stored T-SQL procedures is that you get a syntax check directly. With dynamic SQL, a trivial syntax error may not show up until run time. Even if you test your code carefully, there may be some query, or some variation of a query, that is only run in odd cases and not covered in your test suite.
It has to be admitted that the strength of this argument is somewhat reduced by the fact that T-SQL is not too industrious on reporting semantic errors. Because of deferred name resolution, SQL Server will not examine queries in stored procedures, where one or more tables are missing, be that misspellings or temp tables created within the procedure. Nevertheless, SQL Server does report sufficiently many errors, for this to be a very important reason to use stored procedures.
Another side of this coin is that when you write dynamic SQL, you embed the
SQL code into strings, which makes programming far more complex. Your SQL
code is a string delimited by single quotes(
'), and this string
may include strings itself, and to include a single quote into the string you
need to double it. You can easily get lost in a maze of quotes if you don't
watch out. (In the section Good Coding Practices
and Tips for Dynamic SQL, we will look a little closer
on how to deal
with this problem.) The most commonly used client languages with T-SQL -
Visual Basic, C#, C++, VBScript – all use the double quote (
as their string delimiter, so dynamic SQL in client code or CLR stored
procedures is less prone to that particular problem. Then again, in VB you
don't have multi-line strings, so at the end of each line you have to have a double
quote, an ampersand and an underscore for continuation. It sure does not
serve to make coding easier. You are relieved from all this hassle, if you
use stored procedures with static SQL only.
Somewhat surprisingly, one of the strongest arguments for stored procedures today may be that they permit you to quickly address bugs and performance problems in the application.
Say that you generate SQL statements in your application, and that there is an error in it. Or that it simply performs unbearably slow. To fix it, you need to build a new executable or DLL, which is likely to contain other code that also has changed since the module was shipped. This means that before the fix can be put into production, the module will have to go through QA and testing.
On the other hand, if the problem is in a stored procedure, and the fix is trivial, you may be able to deploy a fix into production within an hour after the problem was reported.
This difference is even more emphasised, if you are an ISV and you ship a product that the customer is supposed administer himself. If your application uses stored procedures, a DBA may be able to address problems directly without opening a support case. For instance, if a procedure runs unacceptably slow, he may be able to fix that by adding an index hint. In contrast, with an application that generates SQL in the client, his hands will be tied. Of course, as an ISV you may not want your customers to poke around in your code, even less to change it. You may also prefer to ship your procedures WITH ENCRYPTION to protect your intellectual property, but this is best controlled through license agreements. (If you encrypt your procedures, the DBA can still change them, as long as he is able to find a way to decrypt them. Which any DBA that knows how to use Google can do.)
In this case, it does not matter whether the stored procedure uses static SQL only, or if it also uses dynamic SQL. For CLR procedures it depends on many objects you have in your assemblies. If you have one assembly per object, installing a new version of a CLR procedure is as simple as replacing a T-SQL procedure.
(I should add that SQL 2005 offers a new feature that permits the DBA to change the plan for a query without altering the code, by adding a plan guide. This feature has been further enhanced in SQL 2008. This is quite an advanced feature, and I refer to Books Online for details.)
Writing dynamic SQL is a task that requires discipline to avoid losing control over your code. If you just go ahead, your code can become very messy, and be difficult to read, troubleshoot and maintain. In this section, we will look at how to avoid this. I will also discuss some special cases: how you can use sp_executesql for input longer than 4000 chars in SQL 2000, and how to use dynamic SQL with cursors, and the combination of dynamic SQL and user-defined functions.
When you write a stored procedure that generates dynamic SQL, you should always include a @debug parameter:
CREATE PROCEDURE dynsql_sp @par1 int, ... @debug bit = 0 AS ... IF @debug = 1 PRINT @sql
When you get a syntax error from the dynamic SQL, it can be very confusing, and you may not even discern where it comes from. And even when you do, it can be very difficult to spot the error only by looking at the code that constructs the SQL. Once the SQL code is slapped in your face, the error is much more likely to be apparent to you. So always include a @debug parameter and a PRINT!
As I've already mentioned, one problem with dynamic SQL is that you often need to deal with nested string delimiters. For instance, in the beginning of this article, I showed you the procedure general_select2. Here it is again:
CREATE PROCEDURE general_select2 @tblname nvarchar(127), @key varchar(10) AS EXEC('SELECT col1, col2, col3 FROM ' + @tblname + ' WHERE keycol = ''' + @key + '''')
(Again, I like to emphasise that this sort of procedure is poor use of dynamic SQL.)
SQL is one of those language where the method to include a string
delimiter itself in a string literal is to double it. So those four consecutive
('''') is a string literal with the value of a one
single quote (
This is a fairly simple example; it can get a lot worse. If you work with
dynamic SQL, you must learn to master nested strings. Obviously, in this case you
can easily escape the mess by using sp_executesql instead – yet another reason
to use parameterised statements. However, there are situations when you need to
deal with nested quotes even with sp_executesql. For instance, earlier in this
article, I had this code:
N' WHERE LastUpdated BETWEEN @fromdate AND ' N' coalesce(@todate, ''99991231'')'
We will look at some tips of dealing with nested strings later in this section.
Another thing to be careful with is the spacing as you concatenate the parts of a query. Here is an example where it goes wrong:
EXEC('SELECT col1, col2, col3 FROM' + @tblname + ' WHERE keycol = ''' + @key + '''')
See that there is a space missing after FROM? When you compile the stored procedure you will get no error, but when you run it, you will be told that the columns keycol, col1, col2, col3 are missing. And since you know that the table you passed to the procedure has these columns you will be mighty confused. But this is the actual code generated, assuming the parameters foo and abc:
SELECT col1, col2, col3 FROMfoo WHERE keycol = 'abc'This is not a syntax error, because
FROMfoois a column alias to col3. And, yes, it's legal to use a WHERE clause, even if there is no FROM clause. But since the columns cannot exist out of the blue, you get an error for that.
This is also a good example why you should use debug prints. If the code looks like this:
SELECT @sql =' SELECT col1, col2, col3 FROM' + @tblname + ' WHERE keycol = ''' + @key + '''' IF @debug = 1 PRINT @sql EXEC(@sql)
It would be much easier to find the error by running the procedure with @debug = 1. (Obviously, had we included the dbo prefix, this error could not occur at all.)
Overall, good formatting is essential when working with dynamic SQL. Try to write the query as you would have written it in static SQL, and then add the string delimiters outside of that. T-SQL permits you to embed newlines in string literals (as testified by the example above), so in contrast to VB, you don't need a string delimiter on each line. An advantage of this is that your debug PRINT is easier to read, and in the case of a syntax error the line number in the error message may guide you.
You may prefer, though, to have a string terminator on each line. A tip in such case is to do something like this:
EXEC(' SELECT col1, col2, col3 ' + ' FROM ' + @tblname + ' WHERE keycol = ''' + @key + '''')
As you see, I have a space after the opening single quote on each line to avoid syntax problems due to missing spaces.
Passing table and column names as parameters to a procedure with dynamic SQL is rarely a good idea for application code. (It can make perfectly sense for admin tasks). As I've said, you cannot pass a table or a column name as a parameter to sp_executesql, but you must interpolate it into the SQL string. Still you should protect it against SQL injection, as a matter of routine. It could be that bad it comes from user input.
To this end, you should use the built-in function quotename() (added in
SQL 7). quotename() takes two parameters: the first is a string, and the second
is a pair of delimiters to wrap the string in. The default for the second
parameter is . Thus,
[Orders]. quotename() takes care of nested delimiters, so if you have
a really crazy table name like Left]Bracket, quotename() will
Note that when you work with names with several components, each component
should be quoted separately.
[dbo.Orders], but that is a table in an unknown
schema of which the first four characters are d, b, o and
a dot. As long as you only work with the dbo schema, best practice is to
add dbo in the dynamic SQL and only pass the table name. If you work
with different schemas, pass the schema as a separate parameter. (Although
you could use the built-in function parsename() to split up a
parameter in parts.)
While general_select still is a poor idea as a stored procedure, here is nevertheless a version that summarises some good coding virtues for dynamic SQL:
CREATE PROCEDURE general_select @tblname nvarchar(128), @key varchar(10), @debug bit = 0 AS DECLARE @sql nvarchar(4000) SET @sql = 'SELECT col1, col2, col3 FROM dbo.' + quotename(@tblname) + ' WHERE keycol = @key' IF @debug = 1 PRINT @sql EXEC sp_executesql @sql, N'@key varchar(10)', @key = @key
The main purpose of quotename() is to quote object names, which is why the default for the second parameter is brackets. But you can specify other delimiters as well, including single quotes, which means that any single quote in the input is doubled. Thus, if you for some reason prefer to use EXEC(), you can use quotename() to protect yourself against SQL injection by help of this function. Here is an example.
IF @custname IS NOT NULL SELECT @sql = @sql + ' AND custname = ' + quotename(@custname, '''')
Say that @custname has the value D'Artagnan. This part of the dynamic SQL becomes:
AND custname = 'D''Artagnan'
There is a limitation with quotename(): its input parameter is nvarchar(128), so it does not handle long strings. A remedy is this user-defined function:
CREATE FUNCTION quotestring(@str nvarchar(1998)) RETURNS nvarchar(4000) AS BEGIN DECLARE @ret nvarchar(4000), @sq char(1) SELECT @sq = '''' SELECT @ret = replace(@str, @sq, @sq + @sq) RETURN(@sq + @ret + @sq) END
This version is for SQL 2000. On SQL 2005 and later, replace 1998 and 4000 with MAX, to make it work for any string length. Here is an example of using this function:
IF @custname IS NOT NULL SELECT @sql = @sql + ' AND custname = ' + dbo.quotestring(@custname)
The result is the same as above.
On SQL 7, you would have to implement quotestring as a stored procedure. SQL 6.5 does not have replace(), so you are a bit out of luck there.
So with quotename() and quotestring(), do we have as good protection against SQL injection as we have with parameterised commands? Maybe. I don't know of any way to inject SQL that slips through quotename() or quotestring(). Nevertheless, you are interpolating user input into the SQL string, whereas with parameterised commands, you don't.
(I should add that I got the suggestion to use quotename() or a user-defined function from SQL Server MVP Steve Kass.)
Another alternative to
escape the mess of nested quotes, is make use
of the fact that T-SQL actually has two string delimiters. To wit, if the
setting QUOTED_IDENTIFIER is OFF, you can also use double quotes(
as a string delimiter. The default
for this setting depends on context, but the preferred setting is
ON, and it
must be ON in order to use XQuery, indexed views and indexes on computed columns.
Thus, this is not a first-rate alternative, but if you are aware of the caveats,
you can do this:
CREATE PROCEDURE general_select @tblname nvarchar(127), @key key_type, @debug bit = 0 AS DECLARE @sql nvarchar(4000) SET @sql = 'SET QUOTED_IDENTIFIER OFF SELECT col1, col2, col3 FROM dbo.' + quotename(@tblname) + ' WHERE keycol = "' + @key + '"' IF @debug = 1 PRINT @sql EXEC(@sql)
Since there are two different quote characters, the code is much easier to read. The single quotes are for the SQL string and the double quotes are for the embedded string literals.
All and all, this is an inferior method to both sp_executesql and quotestring(), since you are not protected against SQL injection (what if @key includes a double quote?). But it would be OK to do for some sysadmin task (where SQL injection is not likely to be an issue), and it may be the best way to go on SQL 6.5.
There is a limitation with sp_executesql on SQL 2000 and SQL 7, since you cannot use longer SQL strings than 4000 characters. (On SQL 2005 and later, you should use nvarchar(MAX) to avoid this problem.) If you want to use sp_executesql when your query string exceeds this limit to make use of parameterised query plans, there is actually a workaround. To wit, you can wrap sp_executesql in EXEC():
DECLARE @sql1 nvarchar(4000), @sql2 nvarchar(4000), @state char(2) SELECT @state = 'CA' SELECT @sql1 = N'SELECT COUNT(*)' SELECT @sql2 = N'FROM dbo.authors WHERE state = @state' EXEC('EXEC sp_executesql N''' + @sql1 + @sql2 + ''', N''@state char(2)'', @state = ''' + @state + '''')
This works, because the @stmt parameter to sp_executesql is ntext, so by itself, it does not have any limitation in size.
You can even use output parameters by using INSERT-EXEC, as in this example:
CREATE TABLE #result (cnt int NOT NULL) DECLARE @sql1 nvarchar(4000), @sql2 nvarchar(4000), @state char(2), @mycnt int SELECT @state = 'CA' SELECT @sql1 = N'SELECT @cnt = COUNT(*)' SELECT @sql2 = N'FROM dbo.authors WHERE state = @state' INSERT #result (cnt) EXEC('DECLARE @cnt int EXEC sp_executesql N''' + @sql1 + @sql2 + ''', N''@state char(2), @cnt int OUTPUT'', @state = ''' + @state + ''', @cnt = @cnt OUTPUT SELECT @cnt') SELECT @mycnt = cnt FROM #result
You have my understanding if you think this is too messy to be worth it.
This very simple: you cannot use dynamic SQL from used-defined functions written in T-SQL. This is because you are not permitted do anything in a UDF that could change the database state (as the UDF may be invoked as part of a query). Since you can do anything from dynamic SQL, including updates, it is obvious why dynamic SQL is not permitted.
I've seen more than one post on the newsgroups where people have been banging their head against this. But if you want to use dynamic SQL in a UDF, back out and redo your design. You have hit a roadblock, and in SQL 2000 there is no way out.
In SQL 2005 and later, you could implement your function as a CLR function. Recall that all data access from the CLR is dynamic SQL. (You are safe-guarded, so that if you perform an update operation from your function, you will get caught.) A word of warning though: data access from scalar UDFs can often give performance problems. If you say
SELECT ... FROM tbl WHERE dbo.MyUdf(somecol) = @value
and MyUdf performs data access, you have more or less created a hidden cursor.
Not that cursors are something you should use very frequently, but people often ask about using dynamic SQL with cursors, so I give an example for the sake of completeness. You cannot say DECLARE CURSOR EXEC(); you have to put the entire DECLARE CURSOR statement in dynamic SQL:
SELECT @sql = 'DECLARE my_cur INSENSITIVE CURSOR FOR ' + 'SELECT col1, col2, col3 FROM ' + @table EXEC sp_executesql @sql
You may be used to using the LOCAL keyword with your cursors. However, it is important to understand that you must use a global cursor, as a local cursor will disappear when the dynamic SQL exits. (Because, as you know by now, the dynamic SQL is its own scope.) Once you have declared the cursor in this way, you can use the cursor in a normal fashion. You must be extra careful with error-handling though, so that you don't exit the procedure without deallocating the cursor.
There is however a way to use locally-scoped cursors with dynamic SQL. Anthony Faull pointed out to me that you can achieve this with cursor variables, as in this example:
DECLARE @my_cur CURSOR EXEC sp_executesql N'SET @my_cur = CURSOR STATIC FOR SELECT name FROM dbo.sysobjects; OPEN @my_cur', N'@my_cur cursor OUTPUT', @my_cur OUTPUT FETCH NEXT FROM @my_cur
You refer to a cursor variable, just like named cursors, but there is an @ in front, and, as you see from the example, you can pass them as a parameters. (I have to confess I have never seen any use for cursor variables until Anthony Faull was kind to send me this example.)
A special feature added in SQL 2005 is that you can use EXEC() to run pass-through queries on a linked server. This could be another instance of SQL Server, but it could also be an Oracle server, an Access database, Active directory or whatever. The SQL could be a single query or a sequence of statements, and could it be composed dynamically or be entirely static. The syntax is simple, as seen by this example:
EXEC('SELECT COUNT(*) FROM ' + @db + '.dbo.sysobjects') AT SQL2K
SQL2K is here a linked server that has been defined with sp_addlinkedserver.
There is one thing that you can do with EXEC() at a linked server, that you
cannot do with EXEC() on a local server: you can use parameters, both for
input and output. The confuse matters, you don't use parameters with names
starting with @, instead you use question marks (
?) as parameter
holders. Say that you are on an SQL 2005 box, and you are dying to know how
many orders VINET had in the Northwind database. Unfortunately, SQL 2005 does
not ship with Northwind, but you have a linked server set up to an instance
of SQL 2000 with Northwind. You can run this:
DECLARE @cnt int EXEC('SELECT ? = COUNT(*) FROM Northwind.dbo.Orders WHERE CustomerID = ?', @cnt OUTPUT, N'VINET') AT SQL2K SELECT @cnt
Note here that the parameter values must appear in the order the parameter markers appear in the query. When passing a parameter, you can either specify a constant value or a variable.
You may ask why the inconsistency with a different parameter marker from
sp_executesql? Recall that linked servers in SQL Server are always accessed
through an OLE DB provider, and OLE DB uses
the parameter marker, a convention inherited from ODBC. OLE DB translates
that parameter marker as is appropriate for the data source on the other end.
(Not all RDBMS use @ for variables.)
As with regular EXEC(), you can specify AS USER/LOGIN to use impersonation:
EXEC('SELECT COUNT(*) FROM ' + @db + '.dbo.sysobjects') AS USER = 'davidson' AT SQL2K
This begs the question: is davidson here a local user or a remote user at SQL2K? Books Online is not very clear about this, but I did some quick experimenting, and found that what you are impersonating is a local user or login, not a login on the remote server. (The login to use on the remote server can be defined with sp_addlinkedsrvlogin.)
When you read the various newsgroups on SQL Server, there is almost every day someone who asks a question that is answered with use dynamic SQL with a quick example to illustrate, but ever so often the person answering forgets to tell about the implications on permissions or SQL injection. On top of that, far too many examples use EXEC() without any thought of query plans. And while many of these questions taken by the letter have no other answer than dynamic SQL, there is often a real business problem which has a completely different solution without dynamic SQL – and a much better one.
So, in this section I will explore some situations where you could use dynamic SQL. You will see that sometimes dynamic SQL is a good choice, but also that in many cases that it is an outright bad idea.
A common question is why the following does not work:
CREATE PROCEDURE my_proc @tablename sysname AS SELECT * FROM @tablename
As we have seen, we can make this procedure work with help of dynamic SQL, but it should also be clear that we gain none of the advantages with generating that dynamic SQL in a stored procedure. You could just as well send the dynamic SQL from the client. So, OK: 1) if the SQL statement is very complex, you save some network traffic and you do encapsulation. 2) As we have seen, starting with SQL 2005 there are methods to deal with permissions. Nevertheless, this is a bad idea.
There seems to be several reasons why people want to parameterise the table name. One camp appears to be people who are new to SQL programming, but have experience from other languages such as C++, VB etc where parameterisation is a good thing. Parameterising the table name to achieve generic code and to increase maintainability seems like good programmer virtue.
But it is just that when it comes to database objects, the old truth does not hold. In a proper database design, each table is unique, as it describes a unique entity. (Or at least it should!) Of course, it is not uncommon to end up with a dozen or more look-up tables that all have an id, a name column and some auditing columns. But they do describe different entities, and their semblance should be regarded as mere chance, and future requirements may make the tables more dissimilar.
Furthermore, when it comes to building a query plan, each table has its set of statistics and presumptions that are by no means interchangeable, as far as SQL Server is concerned. Finally, in a complex data model, it is important to get a grip of what's being used. When you start to pass table and column names as parameters, you definitely lose control.
So if you want to do the above (save the fact that SELECT * should not be used in production code), to save some typing, you are on the wrong path. It is much better to write ten or twenty stored procedures, even if they are similar to each other.
(If your SQL statements are complex, so that there actually is a considerable gain in maintainability to only have them in one place, despite different tables being used, you could consider using a pre-processor like the one in C/C++. You would still have one set of procedures per table, but the code would be in one single include file.)
This is a variation of the previous case, where there is a suite of tables that actually do describe the same entity. All tables have the same columns, and the name includes some partitioning component, typically year and sometimes also month. New tables are created as a new year/month begins.
In this case, writing one stored procedure per table is not really feasible. Not the least, because the user may want to specify a date range for a search, so even with one procedure per table you would still need a dynamic dispatcher.
Now, let's make this very clear: this is a flawed table design. You should not have one sales table per month, you should have one single sales table, and the month that appear in the table name, should be the first column of the primary key in the united sales table. But you may be stuck with a legacy application where you cannot easily change the table design. And, admittedly, there are situations where partitioning makes sense. The table may be huge (say over 10 GB in size), or you want to be able age to out old data quickly. But in such case you should do partitioning properly.
In the following, I will look at three approaches to deal with partitioning without using dynamic SQL.
Partitioned tables were added in SQL 2005. You can divide a table in up to 999 partition according to a partition function. These partitions can be split up over different filegroups to spread out the load. Another important benefit of partitioned tables is that deleting a partition is a pure meta-data operation, which means that if you want to throw away all orders that are more than, say, 12 months old, you can do this with the wink of an eye.
Table partitioning is only available in Enterprise and Developer Edition, not in Standard. For this reason, I'm not going into further details, but refer you to Books Online.
If you have an old application, where you cannot easily merge the umpteen sales tables into one, because it would break other parts of the application, a simple approach is to define a view like this:
CREATE VIEW sales AS SELECT year = '2006', col1, col2, ... FROM dbo.sales2006 UNION ALL SELECT year = '2005', col1, col2, ... FROM dbo.sales2005 UNION ALL ...
Instead of composing the table name dynamically, you can now say:
SELECT ... FROM sales WHERE year = '2006' AND ...
Also, it's easy to add new tables to the view or remove old tables as the data is aged out. Unfortunately, this view is not terribly efficient, as the query will access all tables in the view. Furthermore, the view is not updateable. But with a few more steps, you could make it into what SQL Server knows as a partitioned view, a feature added in SQL 2000 (and available in all editions of SQL Server). A true partitioned view can be very efficient, because for queries that include the partitioning column in the WHERE clause, SQL Server will only access the relevant table(s). And such a view is updatable, so you can insert data into it, and the data will end up in the right table.
Here is a quick example/demo on how to properly set up a partitioned view. Assume that as legacy of a poor design we have these three tables:
SELECT OrderID + 0 AS OrderID, OrderDate, CustomerID, EmployeeID INTO Orders96 FROM Northwind..Orders WHERE year(OrderDate) = 1996 ALTER TABLE Orders96 ALTER COLUMN OrderID int NOT NULL SELECT OrderID + 0 AS OrderID, OrderDate, CustomerID, EmployeeID INTO Orders97 FROM Northwind..Orders WHERE year(OrderDate) = 1997 ALTER TABLE Orders97 ALTER COLUMN OrderID int NOT NULL SELECT OrderID + 0 AS OrderID, OrderDate, CustomerID, EmployeeID INTO Orders98 FROM Northwind..Orders WHERE year(OrderDate) = 1998 ALTER TABLE Orders98 ALTER COLUMN OrderID int NOT NULL go ALTER TABLE Orders97 ADD CONSTRAINT pk97 PRIMARY KEY (OrderID) ALTER TABLE Orders96 ADD CONSTRAINT pk96 PRIMARY KEY (OrderID) ALTER TABLE Orders98 ADD CONSTRAINT pk98 PRIMARY KEY (OrderID)
First step is to a add Year column to each table. These columns need a default (so that processes that insert directly into these tables are unaffected) and a CHECK constraint. Here is how it looks for Orders96:
ALTER TABLE Orders96 ADD Year char(4) NOT NULL CONSTRAINT def96 DEFAULT '1996' CONSTRAINT check96 CHECK (Year = '1996')
This column must be the first column in the primary key, so we need to drop the current primary key and recreate it:
ALTER TABLE Orders96 DROP CONSTRAINT pk96 ALTER TABLE Orders96 ADD CONSTRAINT pk96 PRIMARY KEY (Year, OrderID)
Again, this must be performed for all three tables. Finally, you can create the view:
CREATE VIEW Orders AS SELECT * FROM dbo.Orders96 UNION ALL SELECT * FROM dbo.Orders97 UNION ALL SELECT * FROM dbo.Orders98
Note: I have here use
SELECT * to save some space in the article, but when you
define your real view, you should list the colunms explicitly. There is a risk that columns could come in different
order in the tables.
You now have a proper partitioned view that you can perform inserts and updates through. For instance you can run:
INSERT Orders(Year, OrderID, OrderDate, CustomerID, EmployeeID) VALUES ('1997', 12000, '19970101', 'BERGS', 2)
And if you run a query like:
SELECT OrderID, OrderDate, EmployeeID FROM Orders WHERE Year = @year AND CustomerID = N'BERGS'
SQL Server will at run-time only access the OrdersNN table that maps to @year. If you look at the query plan casually, it may seem that all three tables are accessed, but if you check the Filter operators you will find something called STARTUP EXPR. This means that SQL Server determines at run-time whether to access the table or not. (In fact, when I tested this, I only got this result on SQL 2005 and SQL 2008. On SQL 2000, the start-up expression was not included for some reason I have not been able to understand.)
For your real-world case you may find it prohibitive to change the primary key. In this case you could add a UNIQUE constraint with the partitioning column + the real primary key. This will not be a proper partitioned view, and the view will not be updatable, but with some luck SQL Server may still apply start-up expressions, and access only one of the base tables. At least I got it to work, when I ran a quick test. You should verify that it works for your situation.
When a new table is added with a new year, the view needs to be redefined. If this happens frequently, for instance by each month, you should probably set up a job for this. I leave out example code, but it requires running a cursor over sysobjects to compose a CREATE VIEW statement that you then execute with sp_executesql or EXEC(). That would be an example of good use of dynamic SQL.
This was a concentrated introduction to partitioned views. You can find the full rules for partitioned views under the topic for CREATE VIEW in Books Online. Good reading is also Stefan Delmarco's detailed article SQL Server 2000 Partitioned Views.
If you have very many tables, there is a risk that you will hit a roadblock with a partitioned view: SQL Server only permits 256 tables in a query. Henrik Staun Poulsen suggested an alternate solution that evades this restriction. You first create that new table, with all the data in it. Then you drop the old tables, but replace them with views:
CREATE VIEW sales200612 AS SELECT col1, col2, col3 FROM sales WHERE yearmonth = '200612'
Old functions that uses dynamic SQL or whatever they do, can continue to do so. If they perform INSERT, UPDATE or DELETE operations, you need to implement INSTEAD OF triggers to support this.
Obviously, this solution requires you to produce a lot of code, but you don't have to write it by hand; you can easily write a program in the language of your choice to generate the views and triggers.
In this case people want to update a column which they select at run time. The above is actually legal in T-SQL, but what happens is simply that the variable @colname is assigned the value in @value for each affected row in the table.
In this case dynamic SQL would call for the user to have UPDATE permissions on the table, something not to take lightly. So there is all reason to avoid it. Here is a fairly simple workaround:
UPDATE tbl SET col1 = CASE @colname WHEN 'col1' THEN @value ELSE col1 END, col2 = CASE @colname WHEN 'col2' THEN @value ELSE col2 END, ...
If you don't know about the CASE expression, please look it up in Books Online. It's a very powerful SQL feature.
Then again, one would wonder why people want to do this. Maybe it's because their tables look like this:
CREATE TABLE products (prodid prodid_type NOT NULL, prodname name_type NOT NULL, ... sales_1 money NULL, sales_2 money NULL, ... sales_12 money NULL, PRIMARY KEY (prodid))
It could make more sense to move these sales_n columns to a second table:
CREATE TABLE product_sales (prodid prodid_type NOT NULL, month tinyint NOT NULL, sales money NOT NULL, PRIMARY KEY (prodid, month))
The request here is to determine the name for a column in a result set at run-time. My gut reaction, is that this should be handled client-side. But if your client is a query window is Management Studio or similar, this is kind of difficult. In any case, this is simple to do without any dynamic SQL on SQL 2005 and later:
DECLARE @mycolalias sysname SELECT @mycolalias = 'This week''s alias' CREATE TABLE #temp (a int NOT NULL, b int NOT NULL) INSERT #temp(a, b) SELECT 12, 17 EXEC tempdb..sp_rename '#temp.b', @mycolalias, 'COLUMN' SELECT * FROM #temp
That is, you first get the data into a temp table, and then you use sp_rename to rename the column along your needs. (You need to qualify sp_rename with tempdb to have it to operate in that database.) You will get an informational message Caution: Changing any part of an object name could break scripts and stored procedures, but you may be able to live with that.
This trick works on SQL 2000 too, although not entirely without dynamic SQL: you need put the SELECT from the temp table in EXEC():
EXEC('SELECT * FROM #temp')
This is because on SQL 2000, sp_rename apparently does not trigger a recompile, so if the the SELECT is in the same batch, the statement fails with Invalid column name 'b'. There is yet one thing to be aware of on SQL 2000: you cannot use sp_rename in a stored procedure that is to be run by plain users, as sp_rename thinks you need to be a member of the db_owner or db_ddladmin database roles, even if this is only a temp table. This issue has been addressed in SQL 2005.
In this case the table is in another database which is somehow determined dynamically. There seems to be several reasons why people want to do this, and depending on your underlying reason, the solution is different.
If you for some reason have your application spread over two databases, what you absolutely not should do is to have code that says:
SELECT ... FROM otherdb.dbo.tbl JOIN ...
This is bad, because if someone asks for a second environment on the same server, you have a lot of code to change.
The best solution for this particular problem is to use synonyms, added in SQL 2005:
CREATE SYNONYM otherdbtbl FOR otherdb.dbo.tbl
You can then refer to otherdb.dbo.tbl as just otherdbtbl. If there is a need for a second set of databases, you only have to update the synonyms, and there is no need to use dynamic SQL.
Yet a way to avoid dynamic SQL is to use stored procedures for all inter-database communication. That is, if you are in db1 and need to get data from db2, you call a stored procedure in db2. This can be dynamic, because EXEC permits you to specify a variable that holds the name of the procedure to execute.
SELECT @dbname = quotename(dbname) FROM ... SELECT @sp = @dbname + '..some_sp' EXEC @ret = @sp @par1, @par2...
If you want to get result sets back from db2, look at my article How to Share Data between Stored Procedures for suggestions.
There may still be cases you may find that dynamic SQL is the only feasible situation. This can be done in two ways. The most obvious is:
SELECT @dbname = quotename(dbname) FROM ... SELECT @sql = ' SELECT ... FROM ' + @dbname + ' .dbo.otherdbtbl ' + ' JOIN dbo.localtbl ... ' EXEC sp_executesql @sql, @params, ...
But, if the query is complex, and most of the tables are in the remote database you can also do:
SELECT @sql = ' SELECT ... FROM dbo.othertbl ' + ' JOIN ' + quotename(db_name()) + '.dbo.localtbl ... ' SELECT @dbname = quotename(dbname) FROM ... SELECT @sp_executesql = @dbname + '..sp_executesql' EXEC @sp_executesql @sql, @params, ...
As above, I make use of that you can specify the procedure name dynamically with EXEC. The trick here is that when you specify a system stored procedure in three-part notation with the database name, the procedure executes in the context of that database. Thus, the dynamic SQL in this example runs in @dbname, not the current database.
This sounds to me like some sysadmin venture, and for sysadmin tasks dynamic SQL is usually a fair game, because neither caching nor permissions are issues. Nevertheless there is an kind of alternative: sp_MSforeachdb, demonstrated by this example:
sp_MSforeachdb 'SELECT ''?'', COUNT(*) FROM sysobjects'
As you might guess, sp_MSforeachdb uses dynamic SQL internally, so what you win is that you don't have to write the control loop yourself. I should hasten to add that sp_MSforeachdb is not documented in Books Online, which also means that use of it is not supported by Microsoft and it could be changed or withdrawn from SQL Server without notice.
The scenario here is that you have a suite of databases with identical schema. The typical reason they are different databases and not one, is that every database serves a different customer, and each customer can access his database (but of course no one else's). Some people see a problem with the same stored procedures in fifty databases, and believe that they face a maintenance nightmare. So they get the idea that they should put the procedures in a "master" database. Yes, you can do that. It will give you a much bigger maintenance problem, because your code will entirely littered with dynamic SQL. In fact, if you feel that this is the only alternative, you are better off skipping stored procedures altogether and do all access from client code instead. In such case there is only one place you need to specify the database: the connection string.
What else can you do? Some people might suggest that you should collapse the databases into one, and employ a strict row-level security scheme. Personally, I would never accept such a solution as a potential customer. In a complex application, bugs can easily lead to that information is exposed to people who should not see it. Besides, row-level security cannot be implemented entirely waterproof in SQL Server. Whereas queries only would return the data they should, query plans and error messages may indirectly disclose information to users who are not authorised to see it.
Another wild approach is to use SQL Server's own master database and install the application procedures as system procedures. I have not played with this for a long time, but I am told that it still works in SQL 2008. In any case, this is entirely unsupported. So while I mention the possibility, I don't give you the details on how to do it and I strongly recommend that you don't go there.
What then is the real solution? Install the stored procedures in each database and develop rollout routines for your SQL objects. You need this anyway, the day you want to update the table definitions. This also permits you to have some flexibility. Some customers may prefer to skip an upgrade. Other customers may be prepared to pay for extra functions that only they have access to. Even more importantly, it permits you to easily scale out and move some databases to a second server. I mentioned that as a customer, I would not accept to share database with other customers. In fact, a security-aware customer would not even accept to share the same instance of SQL Server, but require his own instance.
(You may ask whether not synonyms could be used to implement the "master" database. I have not been able to think of anything useful, but if you find out something, please drop me a line.)
This question sometimes comes up. Most often people have problems with the USE command. The correct solution is to avoid USE altogether in this case. In fact, we have already seen how to do this:
SELECT @sql = 'CREATE VIEW ...' SELECT @sp_executesql = quotename(@dbname) + '..sp_executesql' EXEC @sp_executesql @sql
That is, make use of that you can set the database context by calling sp_executesql with three-part notation.
It is fascinating how may people who put '1,2,3,4' in @list, and then are puzzled why the query above does not return any rows. Well, if there is a row where col has the value '1,2,3,4', you will get a match. These two conditions are the same:
col IN (@list) col = @list
IN does not mean "parse whatever data there is at runtime as a
comma-separated list". It's a compile-time shortcut for
@a OR col = @b OR ...
This is a very common question on the newsgroups, and Use dynamic SQL is a far too common answer. Yes, you can do this with dynamic SQL, but it is an extremely poor solution. You cannot pass the list as a parameter to sp_executesql, so you would have to use EXEC() and be open to SQL injection. On top of that, for long lists, IN has extremely poor performance – in some tests I did, it took SQL Server 15 seconds to build the query plan for a list with 10000 elements.
The correct method is to unpack the list into a table with a user-defined function or a stored procedure. In my article, Arrays and Lists in SQL Server, I describe a whole range of ways to do this. I also present performance data for the various methods. (Dynamic SQL is at the bottom of that list!) This is a long article, but there are jump-start links in the beginning of the article, depending on which version of SQL Server you are using.
If you are considering to write the procedure
CREATE PROCEDURE search_sp @condition varchar(8000) AS SELECT * FROM tbl WHERE @condition
Just forget it. If you are doing this, you have not completed the transition to use stored procedure and you are still assembling your SQL code in the client. But this example lapses into
A not too uncommon case is that the users should be able to select data from a broad set of parameters. The procedure search_orders in the section on SQL injection is a very simple example of this.
Any programmer that tackles this realises that writing a static solution with a tailor-made query for each combination of input parameters is impossible. It most cases, it's simple to write a single static query with conditions like:
AND (CustomerID = @custid OR @custid IS NULL)
But in SQL 2005 and earlier if is not possible to get good performance from such a query, but the only option for good performance is to use dynamic SQL. This changed in SQL 2008, provided that you use the RECOMPILE hint. However that is a bit complicated, because the original implementation had a serious bug, so Microsoft reverted on that change for a while. Rather than going into details here, I refer you to my article, Dynamic Search Conditions, where I discuss this type of searches in more detail and where I present several methods, both with dynamic SQL and static SQL. This article exists in two versions, one for SQL 2008 and later, and one for earlier versions.
Another common request is to make a dynamic crosstab query, where you transform rows into columns. For instance, say that you want to display the number of orders handled by each employee in Northwind with one column per year. This query works well:
SELECT E.LastName,  = SUM(CASE Year(OrderDate) WHEN '1996' THEN 1 ELSE 0 END),  = SUM(CASE Year(OrderDate) WHEN '1997' THEN 1 ELSE 0 END),  = SUM(CASE Year(OrderDate) WHEN '1998' THEN 1 ELSE 0 END) FROM Orders O JOIN Employees E ON O.EmployeeID = E.EmployeeID GROUP BY E.LastName
But in many situations you don't exactly which columns you will have in the data, or even how many there will be. For instance, in this example, we may not know beforehand for which years there are orders.
One approach is to set an upper limit of how many output columns you support and use dummy names for the columns. Once you have run the query, you use the technique I described in the section SELECT col AS @myname.
However, in the very most cases, you will want to employ dynamic SQL for this. And there is not really any alternative. A SELECT statement returns a table, and a table has a known number of columns with known names, so there is no way you can write a static SELECT statement to achieve this.
The general technique is a two-step operation: 1) Get the unique values to pivot on. 2) with those values, generate a query like the one above. While it's a short description, it takes some time to get everything in place. You can make a shortcut with the stored procedure pivot_sp, something I have adapted from a procedure originally written by SQL Server MVP Itzik Ben-Gan. The procedure takes a number of parameters permitting you to specify the query, the rows to group by and to pivot by, and which aggregation operation you want.
As this article is already long enough, I don't go into details to try to explain how it works, but leave it to you to explore it on your own. I like to stress one thing though: The way pivot_sp is written, it is wide-open to SQL injection, and it is very difficult, not to say impossible to make the procedure fool-proof since it accepts query text as parameter. This is no problem as long as you use it as your own utility procedure and have full control over the input, but you should not make a procedure like this one accessible to anyone. Rather I recommend that you do
DENY EXECUTE ON pivot_sp TO public
To make sure that plain users cannot run it. If you make a call to pivot_sp in a stored procedure, this call will succeed as ownership chaining applies. The file for pivot_sp includes an example to demonstrate this.
Another option for dynamic crosstab is RAC, which is a third-party tool. I have never used it myself, but I have heard several good comments about it.
This can easily be handled without dynamic SQL in this way:
SELECT col1, col2, col3 FROM dbo.tbl ORDER BY CASE @col1 WHEN 'col1' THEN col1 WHEN 'col2' THEN col2 WHEN 'col3' THEN col3 END
Again, review the CASE expression in Books Online, if you are not acquainted with it.
Note that if the columns have different data types you cannot lump them into the same CASE expression, as the data type of a CASE expression is always one and the same. Instead, you can do this:
SELECT col1, col2, col3 FROM dbo.tbl ORDER BY CASE @col1 WHEN 'col1' THEN col1 ELSE NULL END, CASE @col1 WHEN 'col2' THEN col2 ELSE NULL END, CASE @col1 WHEN 'col3' THEN col3 ELSE NULL END
If you also want to make it dynamic whether the order should be ascending or descending, add one more CASE:
SELECT col1, col2, col3 FROM dbo.tbl ORDER BY CASE @sortorder WHEN 'ASC' THEN CASE @col1 WHEN 'col1' THEN col1 WHEN 'col2' THEN col2 WHEN 'col3' THEN col3 END ELSE NULL END ASC, CASE @sortorder WHEN 'DESC' THEN CASE @col1 WHEN 'col1' THEN col1 WHEN 'col2' THEN col2 WHEN 'col3' THEN col3 END ELSE NULL END
Or use the form in the second example to deal with different data types.
SQL Server MVP Itzik Ben-Gan had a good article on this topic in the March 2001 issue of SQL Server Magazine, where he offers other suggestions.
It should be added that these solutions has the disadvantage that they will always cause a sort which for a large data set could be expensive. If you add an ORDER BY clause in dynamic SQL, the optimizer may avoid the sort if there is a suitable index.
This is no longer an issue, since SQL 2005 added new syntax that permits a variable:
SELECT TOP(@n) col1, col2 FROM tbl
On SQL 2000, TOP does not accept variables, so you need to use dynamic SQL to use TOP. But there is an alternative:
CREATE PROCEDURE get_first_n @n int AS SET ROWCOUNT @n SELECT au_id, au_lname, au_fname FROM authors ORDER BY au_id SET ROWCOUNT 0
It can be disputed whether SET ROWCOUNT @n is really a better solution than running a dynamic SQL statement with TOP. A dynamic TOP is probably a better choice, as long as you can accept the security implications. (But it's not worth to change the permissions only for this.)
I guess a common reason for wanting to do this is to implement paging in web applications. SQL Server MVP Aaron Bertrand has an article which is the standard reference on this topic.
The desire here is to create a table of which the name is determined at run-time.
If we just look at the arguments against using dynamic SQL in stored procedures, few of them are really applicable here. If a stored procedure has a static CREATE TABLE in it, the user who runs the procedure must have permissions to create tables, so dynamic SQL will not change anything. Plan caching obviously has nothing to do with it. Etc.
Nevertheless: Why? Why would you want to do this? If you are creating tables on the fly in your application, you have missed some fundamentals about database design. In a relational database, the set of tables and columns are supposed to be constant. They may change with the installation of new versions, but not during run-time.
Sometimes when people are doing this, it appears that they want to construct unique names for temporary tables. This is completely unnecessary, as this is a built-in feature in SQL Server. If you say:
CREATE TABLE #nisse (a int NOT NULL)
then the actual name behind the scenes will be something much longer, and no other connections will be able to see this instance of #nisse.
If you want to create a permanent table which is unique to a user, but you don't want to stay connected and therefore cannot use temp tables, it may be better to create one table that all clients can share, but where the first column is a key which is private to the client. I discuss this method a little more closely in my article How to Share Data between Stored Procedures.
Sometimes I see persons on the newsgroups that are unhappy, because they create a temp table from dynamic SQL, and then they can't access it, because it disappeared when the dynamic SQL exited. When told that they have to create the table outside the dynamic SQL, they respond that they can't, because they don't know the structure of the table until run-time.
One solution is to create a global temp table, one with two # in the name, for instance ##temp. Such a table is visible to all processes (so you may have to take precautions to make the name unique), and unless you explicitly drop it, it exists until your process exits.
But the real question is: what are these guys up to? If you are working with a relational database, and you don't know the structure of your data until run-time, then there is something fundamentally wrong. As I have never been able to fully understand what the underlying business requirements are, I can't really provide any alternatives. But I would suggest that if you need to go this road, you should seriously consider to run your SQL from a client program. Because, all access to that table would have to be through dynamic SQL, and composing dynamic SQL strings is easier in languages with better string capabilities, be that C#, VB or Perl.
This is similar to parameterising the database name, but in this case we want to access a linked server of which the name is determined at run-time.
Two of the solutions for dynamic database names apply here as well:
CREATE SYNONYM myremotetbl FOR Server.db.dbo.remotetbl
SET @sp = @server + '.db.dbo.some_sp' EXEC @ret = @sp @par1, @par2...
If you want to join a local table with a remote table on some remote server, determined in the flux of the moment, dynamic SQL is probably the best way if you are on SQL 2000. There exists however an alternative, although it's only usable in some situations. You can use sp_addlinkedserver to define the linked server at run-time, as demonstrated by this snippet:
EXEC sp_addlinkedserver MYSRV, @srvproduct='Any', @provider='SQLOLEDB', @datasrc=@@SERVERNAME go CREATE PROCEDURE linksrv_demo_inner WITH RECOMPILE AS SELECT * FROM MYSRV.master.dbo.sysdatabases go EXEC sp_dropserver MYSRV go CREATE PROCEDURE linksrv_demo @server sysname AS IF EXISTS (SELECT * FROM master..sysservers WHERE srvname = 'MYSRV') EXEC sp_dropserver MYSRV EXEC sp_addlinkedserver MYSRV, @srvproduct='Any', @provider='SQLOLEDB', @datasrc=@server EXEC linksrv_demo_inner EXEC sp_dropserver MYSRV go EXEC linksrv_demo 'Server1' EXEC linksrv_demo 'Server2'
There are two procedures. linksrv_demo_inner is the procedure where we actually access the linked server. As the linked server must exist when the procedure is created, I first create a dummy entry for MYSRV, which I subsequently drop once the procedure has been created. (Not only must the linked server exist, it must also have the database and tables that you access.) linksrv_demo is the outside interface which takes a server name as a parameter, and then at run-time defines MYSRV to point to @server.
The above is only possible under certain conditions:
As you can see in the example, I've added WITH RECOMPILE to linksrv_demo_inner. This is a safety precaution, to prevent that a cached plan does not access a different server. I don't think this is really necessary, as SQL Server should sense the changed definition. In fact, you may not even have to split the code over two procedures, but as they say, better safe than sorry.
The rowset functions OPENQUERY and OPENROWSET often calls for dynamic SQL. Their second argument is an SQL string, and they do no accept variables. (This is because the optimizer builds a plan for the distributed query when the procedure is compiled.) So any single parameter you want to pass to the SQL statement for that remote server requires you to use dynamic SQL. Since the remote SQL string can include string literals, you may have to deal with up to three levels of nested quotes. If you don't watch out, you can spend a full day looking at things like:
DECLARE @sql varchar(8000) SELECT @sql = 'SELECT * FROM OPENQUERY(MYSRV, ' + '''SELECT * FROM Northwind.dbo.Orders ' + 'WHERE CustomerID = N''''VINET'''''')' PRINT @sql EXEC(@sql)
to try to find out if you might you have one
' too many or too
Strict discipline is absolutely necessary when working with dynamic SQL for OPENQUERY. The function quotestring() that I showed you earlier can be of great help:
DECLARE @remotesql nvarchar(4000), @localsql nvarchar(4000), @state char(2) SELECT @state = 'CA' SELECT @remotesql = 'SELECT * FROM pubs.dbo.authors WHERE state = ' + dbo.quotestring(@state) SELECT @localsql = 'SELECT * FROM OPENQUERY(MYSRV, ' + dbo.quotestring(@remotesql) + ')' PRINT @localsql EXEC (@localsql)
The built-in function quotename() is usually not useful here, as the SQL statement easily can exceed the limit of 129 characters for the input parameter to quotename().
On SQL 2005 and later, you can use EXEC() to run an SQL statement on a linked server. Since EXEC() at linked servers can take parameters, this can make things considerably easier. Then again, you can join OPENQUERY with local tables, so that only rows of interest are brought across the wire. This you cannot do with EXEC().
Say that you write a stored procedure that is to present some data, and the GUI it is to be run from is Query Analyzer or SQL Server Management Studio (presumably because it is a sysadmin procedure). To make the output easy to digest, you want the column width to be so wide that no data is truncated, but neither do you want any extraneous spaces. This is something you can achieve with dynamic SQL. Typically you would use a temp table to hold the data, in which case there are no permission issues.
Rather than giving an example, I refer you to the source code for the popular (but undocumented) system procedure sp_who2. You can find the code by entering exec master..sp_helptext sp_who2.
I've written this text with a main focus on application code, because it is mainly in application tasks, bad usage of dynamic SQL can cause serious harm by opening for SQL injection, poor query-plan reuse, and result in code that is difficult to read and maintain.
Here, I like to briefly discuss code is for maintenance jobs, code that runs once a night or once a week or even less frequently. Generally, for this sort of code, dynamic SQL is almost always a fair game. Query plans are rarely an issue. And if the code is to be run by users with sysadmin privileges, there are no permissions issues. The same applies to code that does not require permissions outside the database, and is to be run by users with db_owner privileges.
There are however, two points about SQL injection I like to make.
I like to thank the following persons who have provided valuable suggestions and input for this article: SQL Server MVPs Tibor Karaszi, Keith Kratochvil, Steve Kass, Umachandar Jaychandran, Hal Berenson and Aaron Bertrand, as well as Pankul Verma, Anthony Faull, Henrik Staun Poulsen, Karl Jones, Jim Higgins, Marcus Hansfeldt, Gennadi Gretchkosiy, Jeremy Lubich and Simon Hayes. I also like to thank Frank Kalis for providing the German translation.
Not the least I like to thank all you people who have pointed out typos and spelling errors. Just keep those letters and cards coming!
If you have suggestions for improvements or corrections on contents, language or formatting, please mail me at firstname.lastname@example.org. If you have technical questions that any knowledgeable person could answer, I encourage you to post to any of the newsgroups microsoft.public.sqlserver.programming or comp.databases.ms-sqlserver.
2015-04-14 – A very small change, in fact the addition of a single dot that was missing in
SET @sp in the section linked servers. Thanks to Jim Higgins for pointing out this error.
2011-06-23 – Updated the text in the section on Dynamic Search Conditions to mention the RECOMPILE hint, since the bug mentioned in the entry from 2009-02-14 has been fixed.
2009-09-12 – Gennadi Gretchkosiy pointed out using that
SELECT * in a
partitioned view is not OK. If the columns come in different order in the underlying tables, you will get a mess.
2009-02-14 - Removed text in the section on Dynamic Search Conditions that referred to the new behaviour of OPTION (RECOMPILE) in SQL 2008, that Microsoft is now reverting on because of a serious bug.
2008-12-06 – Modified the section on dynamic crosstab to stress that pivot_sp is open to SQL injection.
2008-12-02 – Updated the article for SQL 2008. Rewrote the section on Sales + @yymm tables and added one more approach. Added a section on dynamic crosstab. Various other minor changes.
2008-06-06 – Added an example on how to deal with dynamic sort order in the section on ORDER BY.
2008-03-18 – Added a section on how to handle a dynamic column alias.
2006-12-27 – In the section Caching Query Plans, added a note on forced parameterisation and a demo of the performance penalty for failing to use parameterised queries.
2006-07-25 – Corrected syntax in example with cursor variable after comment from Anthony Faull.
2006-04-23 – Thoroughly reworked the article to cover SQL 2005 in full, resulting in lots of new text, lots of old text dropped, and many sections rearranged. I'm now more strongly favouring sp_executesql over EXEC(), and I put more stress on SQL injection. I also stress the importance of using parameterised statements for query-plan reuse, and I note that prefixing with dbo is essential for query-plan reuse. The examples of cases where (not) to use dynamic SQL have had an overhaul as well, if not equally drastic. I'm now giving a very quick example of partitioned views for the sales + @yymm case. The article now also includes snippets for parameterised commands from VB6 and VB .Net.
2005-04-17 – Added example of EXEC + sp_executesql with OUTPUT parameter. Added use of nvarchar(max) on SQL 2005 for quotestring and elsewhere.
2004-02-08 – German translation now available. Minor language corrections.
2003-12-02 – Added example of using cursor variable with dynamic SQL. Modified description of first parameter to sp_executesql.
Back to my home page.