对于Microsoft SQL Server，您应该尽可能使用存储过程来帮助执行计划缓存和重用。为什么要优化计划重用?因为生成执行计划的成本相当高。

Although the caching and reuse of execution plans for ad-hoc queries has improved significantly in later editions of SQL server (especially 2005 and 2008) there are still far fewer issues with plan reuse when dealing with stored procedures than there are for ad-hoc queries. For example, SQL server will only re-use an execution plan if the plan text matches exactly - right down to comments and white space, for example, if each of the following lines of SQL were to be executed independently, none of them would use the same execution plan:

SELECT MyColumn FROM MyTable WHERE id = @id
select MyColumn from MyTable WHERE id = @id
SELECT MyColumn  FROM MyTable WHERE id = @id
SELECT MyColumn FROM MyTable WHERE id = @id -- "some comment"
SELECT MyColumn FROM MyTable WHERE id = @id -- "some other comment"

除此之外,如果你不显式地指定类型的参数然后有一个好的机会,SQL Server可能出错,例如如果您执行上面的查询与输入4,然后用@ id查询SQL Server将parametrise SMALLINT(或可能是一个非常小的整数),所以如果你然后执行相同的查询@ id的说4000,SQL Server将parametrise INT,而不会重用相同的缓存。

我认为还有其他一些问题，老实说，大多数问题都可以解决——特别是在SQL Server的后续版本中，但是存储过程通常会提供更多的控制。

在某些情况下，在代码中动态创建的sql可能比存储的proc具有更好的性能。如果您已经创建了一个存储的proc(例如sp_customersearch)，它必须非常灵活，因此具有许多参数，非常复杂，那么您可能可以在运行时在代码中生成一个更简单的sql语句。

有人可能会说，这只是将一些处理从SQL转移到web服务器，但总的来说，这是一件好事。

这种技术的另一个优点是，如果你在SQL分析器中查找，你可以看到你生成的查询，调试它比看到一个存储的带有20个参数的proc调用要容易得多。

我站在代码这一边。我们构建了所有应用程序(包括web和客户端)使用的数据访问层，所以从这个角度来看这是DRY。它简化了数据库部署，因为我们只需要确保表模式是正确的。它简化了代码维护，因为我们不需要查看源代码和数据库。

在与数据模型的紧密耦合方面，我没有遇到太多问题，因为我看不出哪里有可能真正打破这种耦合。应用程序及其数据在本质上是耦合的。

CON

我发现在存储过程中进行大量的处理会使您的DB服务器在扩展您的行为时成为一个单一的不灵活点。

然而，如果您有多个服务器运行您的代码，那么在您的程序中进行所有这些处理而不是sql-server，可能会允许您进行更多的扩展。当然，这并不适用于只进行正常获取或更新的存储procs，而是适用于执行更多处理(如在数据集上循环)的存储procs。

PROS

Performance for what it may be worth (avoids query parsing by DB driver / plan recreation etc) Data manipulation is not embedded in the C/C++/C# code which means I have less low level code to look through. SQL is less verbose and easier to look through when listed separately. Due to the separation folks are able to find and reuse SQL code much easier. Its easier to change things when schema changes - you just have to give the same output to the code and it will work just fine Easier to port to a different database. I can list individual permissions on my stored procedures and control access at that level too. I can profile my data query/ persistence code separate from my data transformation code. I can implement changeable conditions in my stored procedure and it would be easy to customize at a customer site. It becomes easier to use some automated tools to convert my schema and statements together rather than when it is embedded inside my code where I would have to hunt them down. Ensuring best practices for data access is easier when you have all your data access code inside a single file - I can check for queries that access the non performant table or that which uses a higher level of serialization or select *'s in the code etc. It becomes easier to find schema changes / data manipulation logic changes when all of it is listed in one file. It becomes easier to do search and replace edits on SQL when they are in the same place e.g. change / add transaction isolation statements for all stored procs. I and the DBA guy find that having a separate SQL file is easier / convenient when the DBA has to review my SQL stuff. Lastly you don't have to worry about SQL injection attacks because some lazy member of your team did not use parametrized queries when using embedded sqls.

我通常写OO代码。我猜你们大多数人可能也有同感。在这种上下文中，所有业务逻辑(包括SQL查询)都属于类定义，这在我看来是显而易见的。分割逻辑(部分驻留在对象模型中，部分驻留在数据库中)并不比将业务逻辑放到用户界面中更好。

关于存储过程的安全性优势，在前面的回答中已经说了很多。它们分为两大类:

1)限制直接访问数据。这在某些情况下确实很重要，当你遇到这样的情况时，存储过程几乎是你唯一的选择。然而，根据我的经验，这样的情况是例外而不是规则。

2) SQL injection/parametrized queries. This objection is a red herring. Inline SQL - even dynamically-generated inline SQL - can be just as fully parametrized as any stored proc and it can be done just as easily in any modern language worth its salt. There is no advantage either way here. ("Lazy developers might not bother with using parameters" is not a valid objection. If you have developers on your team who prefer to just concatenate user data into their SQL instead of using parameters, you first try to educate them, then you fire them if that doesn't work, just like you would with developers who have any other bad, demonstrably detrimental habit.)

我想再投一票赞成使用存储过程(尽管在维护和版本控制时它们会带来麻烦)作为一种限制对底层表的直接访问以获得更好的安全性的方法。