@布莱恩特，
首先，@Backs 为您保存了这篇文章。当我第一次看到它时，我想“该死。如果他不想花时间让他的请求可读，我为什么要打扰”？此外，如果您正在寻找一个编码示例，那么最好创建一些易于使用的测试数据，以使人们更容易帮助您。此外，正如@Felix Pamittan 建议的那样，您还应该发布您的预期回报。
这是发布易于使用的测试数据的一种方法。我还添加了另一个 Policy_ID，以便我可以演示如何对整个表执行此操作，而不仅仅是一个 Policy_ID。
--===== If the test table doesn't already exist, drop it to make reruns in SSMS easier.
-- This is NOT a part of the solution. We're just simulating the original table
-- using a Temp Table.
IF OBJECT_ID('tempdb..#TBL_POL','U') IS NOT NULL
DROP TABLE #TBL_POL
;
--===== Create the test table (technically, a heap because no clustered index)
-- Total SWAG on the data-types because you didn't provide those, either.
CREATE TABLE #TBL_POL
(
Policy_ID INT NOT NULL
,Load_DATE DATE NOT NULL
,ENDORSEMENT# TINYINT NOT NULL
,SEQUENCE TINYINT NOT NULL
,EXTRACTDATE DATE NOT NULL
)
;
--===== Populate the test table
INSERT INTO #TBL_POL
(Policy_ID,Load_DATE,ENDORSEMENT#,SEQUENCE,EXTRACTDATE)
SELECT Policy_ID,Load_DATE,ENDORSEMENT#,SEQUENCE,EXTRACTDATE
FROM (VALUES
--===== Original values provided
(25276,'8/16/2015',0,1,'8/15/2015')
,(25276,'2/13/2016',1,2,'2/12/2016')
,(25276,'9/24/2016',3,4,'9/20/2016')
,(25276,'9/24/2016',3,4,'9/20/2016')
,(25276,'9/24/2016',2,3,'9/20/2016')
--===== Additional values to demo multiple Policy_IDs with
,(12345,'8/16/2015',0,1,'8/15/2015')
,(12345,'9/24/2016',1,5,'2/12/2016')
,(12345,'2/13/2016',1,2,'2/12/2016')
,(12345,'9/24/2016',3,4,'9/20/2016')
,(12345,'9/24/2016',3,4,'9/20/2016')
,(12345,'9/24/2016',2,3,'9/20/2016')
) v (Policy_ID,Load_DATE,ENDORSEMENT#,SEQUENCE,EXTRACTDATE)
;
--===== Show what's in the test table
SELECT *
FROM #TBL_POL
;
如果您希望一次解决多个 Policy_ID 的问题，则以下方法将起作用。
--===== Use a partitioned windowing function to find the latest row
-- for each Policy_ID, ignoring "dupes" in the process.
-- This assumes that the "sequence" column is king of the hill.
WITH cteEnumerate AS
(
SELECT *
,RN = ROW_NUMBER() OVER (PARTITION BY Policy_ID ORDER BY SEQUENCE DESC)
FROM #TBL_POL
)
SELECT Policy_ID,Load_DATE,ENDORSEMENT#,SEQUENCE,EXTRACTDATE
FROM cteEnumerate
WHERE RN = 1
;
如果您只是为此寻找一个 Policy_ID，@ZLK 建议的“TOP 1”方法将起作用，但在上面添加 WHERE 子句也是如此。不确定哪个会更快，但相同的索引会对两者都有帮助。这是带有 WHERE 子句（可以参数化）的解决方案。
--===== Use a partitioned windowing function to find the latest row
-- for each Policy_ID, ignoring "dupes" in the process.
-- This assumes that the "sequence" column is king of the hill.
WITH cteEnumerate AS
(
SELECT *
,RN = ROW_NUMBER() OVER (PARTITION BY Policy_ID ORDER BY SEQUENCE DESC)
FROM #TBL_POL
WHERE Policy_ID = 25276
)
SELECT Policy_ID,Load_DATE,ENDORSEMENT#,SEQUENCE,EXTRACTDATE
FROM cteEnumerate
WHERE RN = 1
;