OFFSET ... FETCH 在高分页值时很慢

Question

这是我的场景：

CREATE TABLE [dbo].[tblSMSSendQueueMain](
    [ID] [int] IDENTITY(1,1) NOT NULL,
    [SendMethod] [int] NOT NULL
 CONSTRAINT [PK_tblSMSSendQueueLog] PRIMARY KEY CLUSTERED 
(
    [ID] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY] TEXTIMAGE_ON [PRIMARY]
GO

CREATE TABLE [dbo].[tblSMSSendQueueMainSendStatus](
    [ID] [bigint] IDENTITY(1,1) NOT NULL,
    [QueueID] [int] NULL,
    [SendStatus] [int] NULL,
    [StatusDate] [datetime] NULL,
    [UserID] [int] NULL,
 CONSTRAINT [PK_tblSMSSendQueueMainSendStatus] PRIMARY KEY CLUSTERED 
(
    [ID] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY]
GO

还有一些索引：

CREATE NONCLUSTERED INDEX [IX_tblSMSSendQueueMainSendStatus_SendStatus_Single] ON [dbo].[tblSMSSendQueueMainSendStatus]
(
    [SendStatus] ASC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
GO

CREATE NONCLUSTERED INDEX [IX_tblSMSSendQueueMain_SendMethod] ON [dbo].[tblSMSSendQueueMain]
(
    [SendMethod] ASC,
    [ID] DESC
)WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, SORT_IN_TEMPDB = OFF, DROP_EXISTING = OFF, ONLINE = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
GO

每个表大约有 13m 行 tblSMSSendQueueMainSendStatus 的 QueueID 列是 tblSMSSendQueueMain 的 ID 列的外键。

服务器有一个 8 核 Xeon CPU 和 8GB RAM。

我在分页计划中使用了 offset 和 fetch，它对于 100k 以下的偏移量来说是完美的，但是当偏移量增加（超过 100k）时，查询响应很慢，大约需要 5 或 6 秒才能运行。

这是我的查询：

SELECT q.ID  
FROM tblSMSSendQueueMain q  
INNER JOIN tblSMSSendQueueMainSendStatus qs  
ON q.ID = qs.QueueID  
WHERE 1 = 1  
AND qs.SendStatus = 5  
AND [SendMethod] = 19  
ORDER BY q.ID desc OFFSET 10 * (1000000 - 1) ROWS  
FETCH NEXT 10 ROWS ONLY 

有人知道我哪里出错了吗？

Answer 1

这很慢的原因是服务器获取正确起始行的唯一方法是读取它之前的每一行。

你最好使用Keyset Pagination。而不是通过起始行号进行分页，而是传入起始键的参数。

要使其工作，您必须返回一个或多个唯一列，并且要使其具有高性能，它们应该被很好地索引。

传入@startingRow作为上一批最高的ID，你可以随心所欲地得到这个。例如。我使用了ORDER BY，所以它将是最后一行，或者您的客户端应用程序将能够从变量中检索它。

SELECT TOP (10)
    q.ID  
FROM tblSMSSendQueueMain q  
INNER JOIN tblSMSSendQueueMainSendStatus qs  
ON q.ID = qs.QueueID  
WHERE 1 = 1  
   AND qs.SendStatus = 5  
   AND q.[SendMethod] = 19
   AND qs.ID > @startingRow    -- drop this line for the first query
ORDER BY qs.ID;

---

我必须说，你的查询有点奇怪。如果外键是q.ID = qs.QueueID，那么如果你只是查询q.ID，你会得到多个相同的结果。我怀疑你实际上只想要q.ID，在这种情况下这是你的唯一密钥：

SELECT TOP (10) DISTINCT
    q.ID  
FROM tblSMSSendQueueMain q  
INNER JOIN tblSMSSendQueueMainSendStatus qs  
ON q.ID = qs.QueueID  
WHERE 1 = 1  
   AND qs.SendStatus = 5  
   AND q.[SendMethod] = 19
   AND q.ID > @startingRow    -- drop this line for the first query
ORDER BY q.ID;

或者，我更喜欢EXISTS/IN，因为它更清楚地说明了要求：

SELECT TOP (10)
    q.ID  
FROM tblSMSSendQueueMain q  
WHERE 1 = 1  
   AND q.[SendMethod] = 19
   AND q.ID IN (
      SELECT qs.QueueID
      FROM tblSMSSendQueueMainSendStatus qs
      WHERE qs.SendStatus = 5
   )
   AND q.ID > @startingRow    -- drop this line for the first query
ORDER BY q.ID;