I think this is a case when a static query is just not the way to go. Dynamically build the list for your in clause, escape your single quotes, and dynamically build SQL. In this case you probably won't see much of a difference with any method due to the small list, but the most efficient method really is to send the SQL exactly as it is written in your post. I think it is a good habit to write it the most efficient way, rather than to do what makes the prettiest code, or consider it bad practice to dynamically build SQL.

I have seen the split functions take longer to execute than the query themselves in many cases where the parameters get large. A stored procedure with table valued parameters in SQL 2008 is the only other option I would consider, although this will probably be slower in your case. TVP will probably only be faster for large lists if you are searching on the primary key of the TVP, because SQL will build a temporary table for the list anyway (if the list is large). You won't know for sure unless you test it.

I have also seen stored procedures that had 500 parameters with default values of null, and having WHERE Column1 IN (@Param1, @Param2, @Param3, ..., @Param500). This caused SQL to build a temp table, do a sort/distinct, and then do a table scan instead of an index seek. That is essentially what you would be doing by parameterizing that query, although on a small enough scale that it won't make a noticeable difference. I highly recommend against having NULL in your IN lists, as if that gets changed to a NOT IN it will not act as intended. You could dynamically build the parameter list, but the only obvious thing that you would gain is that the objects would escape the single quotes for you. That approach is also slightly slower on the application end since the objects have to parse the query to find the parameters. It may or may not be faster on SQL, as parameterized queries call sp_prepare, sp_execute for as many times you execute the query, followed by sp_unprepare.

重用存储过程或参数化查询的执行计划可能会提高性能，但它会将您锁定在由执行的第一个查询决定的执行计划中。在许多情况下，这对于后续查询可能不太理想。在您的情况下，重用执行计划可能是一个加分项，但它可能根本没有任何区别，因为示例是一个非常简单的查询。

悬崖笔记:

对于您的情况，您所做的任何事情，无论是使用列表中固定数量的项进行参数化(如果不使用则为空)，动态地构建带有或不带有参数的查询，还是使用带有表值参数的存储过程，都不会产生太大的区别。不过，我的一般建议如下:

你的case/简单查询很少参数:

动态SQL，如果测试显示更好的性能，可能会使用参数。

具有可重用执行计划的查询，通过简单地更改参数或如果查询很复杂则调用多次:

带有动态参数的SQL。

带有大列表的查询:

具有表值参数的存储过程。如果列表变化很大，则在存储过程上使用WITH RECOMPILE，或者简单地使用不带参数的动态SQL为每个查询生成新的执行计划。

这是这个问题的另一个答案。

(新版本发布于6/4/13)。

    private static DataSet GetDataSet(SqlConnectionStringBuilder scsb, string strSql, params object[] pars)
    {
        var ds = new DataSet();
        using (var sqlConn = new SqlConnection(scsb.ConnectionString))
        {
            var sqlParameters = new List<SqlParameter>();
            var replacementStrings = new Dictionary<string, string>();
            if (pars != null)
            {
                for (int i = 0; i < pars.Length; i++)
                {
                    if (pars[i] is IEnumerable<object>)
                    {
                        List<object> enumerable = (pars[i] as IEnumerable<object>).ToList();
                        replacementStrings.Add("@" + i, String.Join(",", enumerable.Select((value, pos) => String.Format("@_{0}_{1}", i, pos))));
                        sqlParameters.AddRange(enumerable.Select((value, pos) => new SqlParameter(String.Format("@_{0}_{1}", i, pos), value ?? DBNull.Value)).ToArray());
                    }
                    else
                    {
                        sqlParameters.Add(new SqlParameter(String.Format("@{0}", i), pars[i] ?? DBNull.Value));
                    }
                }
            }
            strSql = replacementStrings.Aggregate(strSql, (current, replacementString) => current.Replace(replacementString.Key, replacementString.Value));
            using (var sqlCommand = new SqlCommand(strSql, sqlConn))
            {
                if (pars != null)
                {
                    sqlCommand.Parameters.AddRange(sqlParameters.ToArray());
                }
                else
                {
                    //Fail-safe, just in case a user intends to pass a single null parameter
                    sqlCommand.Parameters.Add(new SqlParameter("@0", DBNull.Value));
                }
                using (var sqlDataAdapter = new SqlDataAdapter(sqlCommand))
                {
                    sqlDataAdapter.Fill(ds);
                }
            }
        }
        return ds;
    }

欢呼。

在默认情况下，我将通过向IN条件传递一个表值函数(从字符串返回一个表)来实现这一点。

下面是UDF的代码(我从Stack Overflow的某个地方得到了它，我现在找不到源代码)

CREATE FUNCTION [dbo].[Split] (@sep char(1), @s varchar(8000))
RETURNS table
AS
RETURN (
    WITH Pieces(pn, start, stop) AS (
      SELECT 1, 1, CHARINDEX(@sep, @s)
      UNION ALL
      SELECT pn + 1, stop + 1, CHARINDEX(@sep, @s, stop + 1)
      FROM Pieces
      WHERE stop > 0
    )
    SELECT 
      SUBSTRING(@s, start, CASE WHEN stop > 0 THEN stop-start ELSE 512 END) AS s
    FROM Pieces
  )

一旦你得到了这个，你的代码就会像这样简单:

select * from Tags 
where Name in (select s from dbo.split(';','ruby;rails;scruffy;rubyonrails'))
order by Count desc

除非你有一个长得离谱的字符串，否则这应该与表索引一起工作得很好。

如果需要，你可以把它插入一个临时表，索引它，然后运行一个连接…

步骤1:-

string[] Ids = new string[] { "3", "6", "14" };
string IdsSP = string.Format("'|{0}|'", string.Join("|", Ids));

步骤2:-

@CurrentShipmentStatusIdArray [nvarchar](255) = NULL

步骤3:-

Where @CurrentShipmentStatusIdArray is null or @CurrentShipmentStatusIdArray LIKE '%|' + convert(nvarchar,Shipments.CurrentShipmentStatusId) + '|%'

or

Where @CurrentShipmentStatusIdArray is null or @CurrentShipmentStatusIdArray LIKE '%|' + Shipments.CurrentShipmentStatusId+ '|%'

(编辑:如果表值参数不可用) 最好的方法似乎是将大量的IN参数分割为多个固定长度的查询，这样您就有了许多具有固定参数计数的已知SQL语句，并且没有虚值/重复值，也没有对字符串、XML等进行解析。

下面是我用c#写的一些关于这个主题的代码:

public static T[][] SplitSqlValues<T>(IEnumerable<T> values)
{
    var sizes = new int[] { 1000, 500, 250, 125, 63, 32, 16, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1 };
    int processed = 0;
    int currSizeIdx = sizes.Length - 1; /* start with last (smallest) */
    var splitLists = new List<T[]>();

    var valuesDistSort = values.Distinct().ToList(); /* remove redundant */
    valuesDistSort.Sort();
    int totalValues = valuesDistSort.Count;

    while (totalValues > sizes[currSizeIdx] && currSizeIdx > 0)
    currSizeIdx--; /* bigger size, by array pos. */

    while (processed < totalValues)
    {
        while (totalValues - processed < sizes[currSizeIdx]) 
            currSizeIdx++; /* smaller size, by array pos. */
        var partList = new T[sizes[currSizeIdx]];
        valuesDistSort.CopyTo(processed, partList, 0, sizes[currSizeIdx]);
        splitLists.Add(partList);
        processed += sizes[currSizeIdx];
    }
    return splitLists.ToArray();
}

(你可能有进一步的想法，省略排序，使用valuesDistSort.Skip(processed). take (size[…])而不是list/array CopyTo)。

当插入参数变量时，您可以创建如下内容:

foreach(int[] partList in splitLists)
{
    /* here: question mark for param variable, use named/numbered params if required */
    string sql = "select * from Items where Id in("
        + string.Join(",", partList.Select(p => "?")) 
        + ")"; /* comma separated ?, one for each partList entry */

    /* create command with sql string, set parameters, execute, merge results */
}

我观察过NHibernate对象关系映射器生成的SQL(当查询数据并从中创建对象时)，它在多个查询下看起来最好。在NHibernate中，可以指定批处理大小;如果需要获取许多对象数据行，它将尝试检索与批处理大小相等的行数

SELECT * FROM MyTable WHERE Id IN (@p1, @p2, @p3, ... , @p[batch-size])

，而不是发送数百或数千

SELECT * FROM MyTable WHERE Id=@id

当剩余的id小于批处理大小，但仍然大于一个时，它会分割成更小的语句，但仍然具有一定的长度。

如果批处理大小为100，查询有118个参数，它将创建3个查询:

一个有100个参数(批量大小)， 然后是12个 另一个是6，

但没有一个是118或18。通过这种方式，它将可能的SQL语句限制为可能的已知语句，防止太多不同的查询计划，从而填充缓存，并且大部分永远不会被重用。上面的代码做了同样的事情，但是长度为1000、500、250、125、63、32、16、10到1。超过1000个元素的参数列表也会被分割，以防止由于大小限制而导致的数据库错误。

无论如何，最好有一个直接发送参数化SQL的数据库接口，而不需要单独的Prepare语句和句柄来调用。像SQL Server和Oracle这样的数据库通过字符串相等来记住SQL(值会改变，绑定SQL中的参数不会!)并重用查询计划(如果可用的话)。不需要单独的prepare语句，也不需要在代码中维护查询句柄!ADO。NET是这样工作的，但是Java似乎仍然使用prepare/execute by句柄(不确定)。

关于这个主题，我有自己的问题，最初建议用重复的IN子句填充，但后来更喜欢NHibernate样式语句split: 参数化SQL -在/不在与固定数量的参数，查询计划缓存优化?

这个问题仍然很有趣，即使在被问了5年多之后……

EDIT: I noted that IN queries with many values (like 250 or more) still tend to be slow, in the given case, on SQL Server. While I expected the DB to create a kind of temporary table internally and join against it, it seemed like it only repeated the single value SELECT expression n-times. Time was up to about 200ms per query - even worse than joining the original IDs retrieval SELECT against the other, related tables.. Also, there were some 10 to 15 CPU units in SQL Server Profiler, something unusual for repeated execution of the same parameterized queries, suggesting that new query plans were created on repeated calls. Maybe ad-hoc like individual queries are not worse at all. I had to compare these queries to non-split queries with changing sizes for a final conclusion, but for now, it seems like long IN clauses should be avoided anyway.

另一种可能的解决方案是，不向存储过程传递可变数量的参数，而是传递一个包含您要查找的名称的字符串，但通过将它们围绕在'<>'中使它们唯一。然后使用PATINDEX查找名称:

SELECT * 
FROM Tags 
WHERE PATINDEX('%<' + Name + '>%','<jo>,<john>,<scruffy>,<rubyonrails>') > 0