Есть несколько способов сделать это. Вот один из способов
Declare @table as table(
id int,
rate decimal(10,5) ,
uninteresting varchar(10) ,
updated_on date,
version int )
INSERT INTO @table
VALUES
(123 , 1.20 , 'foo ' , '2010-10-18' , 1500),
(456, 2.1, ' bar ', ' 2010-10-12 ', 2123),
(123, 1.2, ' baz ', ' 2010-10-10 ', 1499),
(123, 1.1, ' baz ', ' 2010-10-08 ', 1498),
(456, 2, ' bar ', ' 2010-10-11 ', 2122),
(123, 1, ' baz ', ' 2010-08-01 ', 1497),
(456, 2, ' quux ', ' 2010-10-05 ', 2121),
(456, 1.95, ' quux ', ' 2010-09-07 ', 2120)
;WITH rates
AS (SELECT Row_number() OVER ( PARTITION BY curr.id, curr.rate ORDER BY curr.updated_on DESC) AS rn,
curr.id,
curr.rate cur_rate,
curr.version cur_ver,
curr.updated_on,
previous.rate prev_rate,
previous.version prev_ver,
previous.updated_on prev_updated
FROM
@table curr
LEFT JOIN @table previous
ON curr.id = previous.id
AND curr.rate <> previous.rate
AND curr.updated_on > previous.updated_on
)
SELECT
id,
cur_rate,
cur_ver,
updated_on,
prev_rate,
prev_ver,
prev_updated
FROM
rates
WHERE
rn = 1
дает этот результат
id cur_rate cur_ver updated_on prev_rate prev_ver prev_updated
----------- -------- ----------- ---------- --------- ----------- ------------
123 1.00000 1497 2010-08-01 NULL NULL NULL
123 1.10000 1498 2010-10-08 1.00000 1497 2010-08-01
123 1.20000 1500 2010-10-18 1.10000 1498 2010-10-08
456 1.95000 2120 2010-09-07 NULL NULL NULL
456 2.00000 2122 2010-10-11 1.95000 2120 2010-09-07
456 2.10000 2123 2010-10-12 2.00000 2122 2010-10-11
ЕСЛИ вы измените значение rn, чтобы снизить скорость в разделе, например,
( PARTITION BY curr.id ORDER BY curr.updated_on DESC) AS rn,
вы получите
id cur_rate cur_ver updated_on prev_rate prev_ver prev_updated
----------- -------- ----------- ---------- --------- ----------- ------------
123 1.20000 1500 2010-10-18 1.10000 1498 2010-10-08
456 2.10000 2123 2010-10-12 2.00000 2122 2010-10-11