使用C#在SQL Server上的临时表中插入3万行的最快方法

Question

我试图找出如何使用c#在SQL Server的临时表中提高插入性能.有些人说我应该使用SQLBulkCopy但是我必须做错了,因为它似乎比简单地构建一个SQL插入字符串慢得多.

我使用SQLBulkCopy创建表的代码如下:

public void MakeTable(string tableName, List<string> ids, SqlConnection connection)
    {

        SqlCommand cmd = new SqlCommand("CREATE TABLE ##" + tableName + " (ID int)", connection);
        cmd.ExecuteNonQuery();

        DataTable localTempTable = new DataTable(tableName);

        DataColumn id = new DataColumn();
        id.DataType = System.Type.GetType("System.Int32");
        id.ColumnName = "ID";
        localTempTable.Columns.Add(id);

        foreach (var item in ids)
        {
             DataRow row = localTempTable.NewRow();
             row[0] = item;
             localTempTable.Rows.Add(row);
             localTempTable.AcceptChanges();
        }


        using (SqlBulkCopy bulkCopy = new SqlBulkCopy(connection))
        {
            bulkCopy.DestinationTableName = "##" + tableName;
            bulkCopy.WriteToServer(localTempTable);

        }
    }

这样我的插入需要很长时间才能运行.我让插件以另一种方式更快地工作:

我将insert位创建为一个字符串,并在我的SQL create temp table语句中加入它:

创建插入字符串:

public string prepareInserts(string tableName, List<string> ids)
    {
        List<string> inserts = new List<string>();

        var total = ids.Select(p => p).Count();
        var size = 1000;

        var insert = 1;

        var skip = size * (insert - 1);

        var canPage = skip < total;

        while (canPage)
        {
            inserts.Add(" insert into ##" + tableName + @" (ID) values " + String.Join(",", ids.Select(p => string.Format("({0})", p))
                        .Skip(skip)
                        .Take(size)
                        .ToArray()));
            insert++;
            skip = size * (insert - 1);
            canPage = skip < total;
        }

        string joinedInserts = String.Join("\r\n", inserts.ToArray());

        return joinedInserts;

    }

创建查询后在SQL语句中使用它们:

inserts = prepareInserts(tableName, ids);

var query = @"IF EXISTS
                                            (
                                            SELECT *
                                            FROM tempdb.dbo.sysobjects
                                            WHERE ID = OBJECT_ID(N'tempdb..##" + tableName + @"')
                                            )
                                                BEGIN
                                                    DELETE FROM ##" + tableName + @"
                                                END
                                            ELSE
                                                BEGIN
                                                    CREATE TABLE ##" + tableName + @"
                                                    (ID int)
                                                END " + inserts;

            var command = new SqlCommand(query, sqlConnection);
...

自从我看到有人告诉我(在堆栈交换机上https://dba.stackexchange.com/questions/44217/fastest-way-to-insert-30-thousand-rows-in-sql-server/44222?noredirect= 1#comment78137_44222)我应该使用SQLBulkCopy,这会更快我相信我应该改进我的方式.因此,如果有人可以建议我如何改进我的SQLBulkCopy代码或告诉我是否有更好的插入语句可以提高我的应用程序的性能,这将是伟大的.

Answer 1

您的问题可能在于,localTempTable.AcceptChanges();因为它提交了您的更改.
如果你做下一个,我认为它会运行得更快

    foreach (var item in ids)
    {
         DataRow row = localTempTable.NewRow();
         row[0] = item;
         localTempTable.Rows.Add(row);

    }

    localTempTable.AcceptChanges();

    using (SqlBulkCopy bulkCopy = new SqlBulkCopy(connection))
    {
        bulkCopy.DestinationTableName = "##" + tableName;
        bulkCopy.WriteToServer(localTempTable);

    }

从MSDN - DataSet.AcceptChanges

提交自加载或自上次调用AcceptChanges以来对此DataSet所做的所有更改.

Answer 2

我自己使用 StopWatch 对象运行此代码来测量时间。每次迭代中的 AcceptChanges 都会导致速度变慢。

public void MakeTable(string tableName, List<string> ids, SqlConnection connection)
{
    SqlCommand cmd = new SqlCommand("CREATE TABLE ##" + tableName + " (ID int)", connection);
    cmd.ExecuteNonQuery();

    DataTable localTempTable = new DataTable(tableName);

    DataColumn id = new DataColumn();
    id.DataType = System.Type.GetType("System.Int32");
    id.ColumnName = "ID";
    localTempTable.Columns.Add(id);

    System.Diagnostics.Stopwatch sw1 = new System.Diagnostics.Stopwatch();        

    sw1.Start();
    foreach (var item in ids)
    {
        DataRow row = localTempTable.NewRow();
        row[0] = item;
        localTempTable.Rows.Add(row);

    }
    localTempTable.AcceptChanges();
    long temp1 = sw1.ElapsedMilliseconds;
    sw1.Reset();
    using (SqlBulkCopy bulkCopy = new SqlBulkCopy(connection))
    {
        bulkCopy.DestinationTableName = "##" + tableName;
        bulkCopy.WriteToServer(localTempTable);

    }
    long temp2 = sw1.ElapsedMilliseconds;
}

当 AccpetChanges 在 foreach 循环内时的结果

当它不是

差异是 3 个数量级 :)