合并表SQL Server中多行的记录

Question

我在t_resourcetable有一些肮脏的资源使用记录，看起来像这样

resNo   subres    startdate                        enddate
1        2        2012-01-02 22:03:00.000          2012-01-03 00:00:00.000
1        2        2012-01-03 00:00:00.000          2012-01-04 00:00:00.000
1        2        2012-01-04 00:00:00.000          2012-01-04 16:23:00.000
1        3        2012-01-06 16:23:00.000          2012-01-06 22:23:00.000
2        2        2012-01-04 05:23:00.000          2012-01-06 16:23:00.000

我需要以这种方式合并那些脏行

resNo   subres    startdate                        enddate
1        2        2012-01-02 22:03:00.000          2012-01-04 16:23:00.000
1        3        2012-01-06 16:23:00.000          2012-01-06 22:23:00.000
2        2        2012-01-04 05:23:00.000          2012-01-06 16:23:00.000

这应该更新到同一张表。 我有超过40k的行，因此不能使用游标。 请帮助我通过更优化的sql语句清理此问题。

提供的解决方案不会遇到类似情况

resNo   subres    startdate                        enddate
1        2        2012-01-02 22:03:00.000          2012-01-03 00:00:00.000
1        2        2012-01-03 00:00:00.000          2012-01-04 00:00:00.000
1        2        2012-01-04 00:00:00.000          2012-01-04 16:23:00.000
1        2        2012-01-14 10:09:00.000          2012-01-15 00:00:00.000
1        2        2012-01-15 00:00:00.000          2012-01-16 00:00:00.000
1        2        2012-01-16 00:00:00.000          2012-01-16 03:00:00.000
1        3        2012-01-06 16:23:00.000          2012-01-06 22:23:00.000
2        2        2012-01-04 05:23:00.000          2012-01-06 16:23:00.000

我需要以这种方式合并那些脏行

resNo   subres    startdate                        enddate
1        2        2012-01-02 22:03:00.000          2012-01-04 16:23:00.000
1        2        2012-01-14 10:09:00.000          2012-01-16 03:00:00.000
1        3        2012-01-06 16:23:00.000          2012-01-06 22:23:00.000
2        2        2012-01-04 05:23:00.000          2012-01-06 16:23:00.000

请协助我解决这个肮脏的数据问题。

Answer 1

MERGE INTO t_resourcetable AS TARGET
USING (
    SELECT
        resNo, subres,
        MIN(startdate) as startdate,
        MAX(enddate) as enddate
    FROM t_resourcetable
    GROUP BY resNo, subres
) AS SOURCE
ON TARGET.resNo = SOURCE.resNo
AND TARGET.subres = SOURCE.subres
AND TARGET.startdate = SOURCE.startdate
-- Set enddate on the first record in the group
WHEN MATCHED THEN
    UPDATE SET TARGET.enddate = SOURCE.enddate
-- Delete the remaining items
WHEN NOT MATCHED BY SOURCE THEN
    DELETE;

---

编辑：要尊重间隔中的差距：

MERGE INTO t_resourcetable AS TARGET
USING (
    -- Find the first item in each interval group
    SELECT
        resNo, subres, startdate,
        row_number() over (partition by resNo, subres order by startdate) as rn
    FROM t_resourcetable t1
    WHERE NOT EXISTS (
        -- No other intervals that intersect this from behind
        SELECT NULL
        FROM t_resourcetable t2
        WHERE t2.resNo = t1.resNo
        AND t2.subres = t1.subres
        AND t2.startdate < t1.startdate
        AND t2.enddate >= t1.startdate
    )
) AS SOURCE_start
INNER JOIN (
    -- Find the last item in each interval group
    SELECT
        resNo, subres, enddate,
        row_number() over (partition by resNo, subres order by startdate) as rn
    FROM t_resourcetable t1
    WHERE NOT EXISTS (
        -- No other intervals that intersect this from ahead
        SELECT NULL
        FROM t_resourcetable t2
        WHERE t2.resNo = t1.resNo
        AND t2.subres = t1.subres
        AND t2.startdate <= t1.enddate
        AND t2.enddate > t1.enddate
    )
) AS SOURCE_end
    ON SOURCE_start.resNo = SOURCE_end.resNo
    AND SOURCE_start.subres = SOURCE_end.subres
    AND SOURCE_start.rn = SOURCE_end.rn -- Join by row number
ON TARGET.resNo = SOURCE_start.resNo
AND TARGET.subres = SOURCE_start.subres
AND TARGET.startdate = SOURCE_start.startdate
-- Set enddate on the first record in the group
WHEN MATCHED THEN
    UPDATE SET TARGET.enddate = SOURCE_end.enddate
-- Delete the remaining items
WHEN NOT MATCHED BY SOURCE THEN
    DELETE;

结果：

resNo   subres   startdate          enddate
    1        2   2012-01-02 22:03   2012-01-04 16:23
    1        2   2012-01-14 10:09   2012-01-16 03:00
    1        3   2012-01-06 16:23   2012-01-06 22:23
    2        2   2012-01-04 05:23   2012-01-06 16:23

---

编辑：如果在目标表上存在并发编辑的任何风险，则可能要添加HOLDLOCK提示。 这将防止出现任何主键冲突错误，并稍微节省资源。 （感谢乔伊）：

MERGE INTO t_resourcetable WITH (HOLDLOCK) AS TARGET
...

Answer 2

对于SQL Server 2005，您可以执行以下操作：

create table #temp
(
  resNo int,
  subres int,
  enddate datetime,
  primary key (resNo, subres)
)

-- Store the values you need for enddate in a temp table
insert into #temp
select resNo, 
       subres,
       max(enddate) as enddate
from t_resourcetable
group by resNo, subres

-- Delete duplicates keeping the row with min startdate
delete T
from (
        select row_number() over(partition by resNo, subres order by startdate) as rn
        from t_resourcetable
     ) as T
where rn > 1

-- Set enddate where needed
update T set enddate = tmp.enddate
from t_resourcetable as T
  inner join #temp as tmp
    on T.resNo = tmp.resNo and
       t.subres = tmp.subres
where T.enddate <> tmp.enddate

drop table #temp

Answer 3

您可以先将结果存储在这样的临时表中：

DECLARE @tmp TABLE
(
    resNo INT, 
    subres INT, 
    startdate DATETIME, 
    enddate DATETIME
)

INSERT   @tmp
SELECT   resNo, subres, MIN(startdate), MAX(enddate)
FROM     t_resourcetable
GROUP BY resNo, subres

要更新t_resourcetable表，您可以执行以下操作：

DELETE   t_resourcetable

INSERT   t_resourcetable
SELECT   * 
FROM     @tmp

并在一个事务中运行所有这些。

Answer 4

我将创建一个临时表。 现在，您可以用新的和清除的数据填充临时表。 我认为，您必须使用resNo和subres组合键，然后选择min startdate和max enddate。

至少，删除旧表中的所有数据，并用临时表中的数据填充它。