Question

我有一个非常大的（100,000,000多行），性能不佳的表格采用以下格式

TABLE [dbo].[Emails](
    [pkid] [int] IDENTITY(1,1) NOT NULL,
    [cid] [int] NULL,
    [email] [nvarchar](255) NULL,
    [lid] [int] NULL,
    [date] [smalldatetime] NULL
)

[{1}}上有一个聚集索引，pkid + cid + email上的唯一非聚集索引，以及非唯一的非聚集索引仅lid]

对于给定的cid，可以有多个条目具有相同的cid和不同的email值。

lid

对于给定的cid, email, lid, date 123, mal@serenity.fake, 456, 2014-07-17 12:21:00 123, mal@serenity.fake, 459, 2014-07-17 12:26:00 123, mal@serenity.fake, 466, 2014-07-17 12:27:00 123, zoe@serenity.fake, 456, 2014-07-17 12:21:00 123, zoe@serenity.fake, 467, 2014-07-17 12:28:00，我想提取前10个唯一的电子邮件，并将其余数据连接到一个lid列表以及一个日期列表或一个最小日期字段。我还想返回所选最后一个数据行的cid，用于分页

所以例如

pkid

最有效的方法是什么？目前实施这种方式似乎非常低效，我想改进它：

mal@serenity.fake, [list of lids], [list of dates], [last pkid for all rows]
zoe@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
wash@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
jane@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
kaylee@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
inara@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
book@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
simon@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
river@serenity.fake, [list of lids], [list of dates], [last pkid for all rows] 
early@intruder.fake, [list of lids], [list of dates], [last pkid for all rows]

正如您所看到的，如果我想要查看这些结果的第5页，它会选择所有5个页面（50封电子邮件，可能是500多行数据），然后从底部选择10个。必须有更好的方法来做到这一点！

Answer 1

这样的事情怎么样？我没有为此编制一个百万行测试表，但执行计划更简单。构建CSV会受到伤害，因为无论你做什么，它都会强制进行第二次表扫描。

with MyRows as
(
    SELECT TOP (@rows * @page) 
        email,
        MIN([date]) as [date],
        COUNT(lid) as [lids],
        [ids] = STUFF(
            (
                SELECT ',' + CAST(lid AS VARCHAR(8)) + ';' + CAST([date] as VARCHAR(20))
                FROM Emails WITH (NOLOCK)
                WHERE cid = @cid 
                AND email = Z.email 
                FOR XML PATH('')
            ), 1, 1, '')
        , ROW_NUMBER() over(PARTITION BY email order by MIN([date]) desc) as RowNum
        FROM Emails Z WITH (NOLOCK) 
        WHERE cid = @cid 
        GROUP BY email 
        ORDER by MIN([date]) DESC 
)

select email
    , [date]
    , lids
    , ids
from MyRows
where RowNum >= @rows * @page 
and RowNum < @rows * (@page + 1)

选择具有10个不同值的所有行

1 个答案: