【发布时间】:2011-05-19 17:42:53
【问题描述】:
我一直想知道在某个外键的历史表中找到最后一个条目的正确方法是什么?
例子:
我们假设我们有一个这样的表:
tHistory(ID | FK_User | State | Timestamp)
Oracle 中读取每个 FK_User ID 的最新条目的正确方法是什么?我总是为此使用子选择,但我觉得它不正确……还有其他方法吗?
【问题讨论】:
我一直想知道在某个外键的历史表中找到最后一个条目的正确方法是什么?
例子:
我们假设我们有一个这样的表:
tHistory(ID | FK_User | State | Timestamp)
Oracle 中读取每个 FK_User ID 的最新条目的正确方法是什么?我总是为此使用子选择,但我觉得它不正确……还有其他方法吗?
【问题讨论】:
另一种方法是使用相关子查询来确定最后日期...
SELECT
*
FROM
tHistory
WHERE
timestamp = (
SELECT
MAX(timestamp)
FROM
tHistory latest
WHERE
FK_User = tHistory.FK_User
)
【讨论】:
SELECT *
FROM (
SELECT h.*,
ROW_NUMBER() OVER
(PARTITION BY h.FK_User ORDER BY timestamp DESC) AS rn
FROM tHistory h
)
WHERE rn = 1
【讨论】:
scalar = (SELECT …)) 与内联视图 (SELECT FROM (SELECT …)) 混淆吗?
好问题。
在版本 8i 之前,您需要使用子查询,为此您需要访问 thistory 表两次。 引入分析函数后,您可以使用内联视图中的 ROW_NUMBER 分析函数跳过第二个表访问,就像 Quassnoi 向您展示的那样。
但是,从第 9 版开始,Oracle 中有一种更好、性能更高的方法。 只需按 FK_USER 分组并使用聚合函数 FIRST 或 LAST。
第一:http://download.oracle.com/docs/cd/B10501_01/server.920/a96540/functions45a.htm#SQLRF00641 最后:http://download.oracle.com/docs/cd/B10501_01/server.920/a96540/functions57a.htm#83735
这是一个带有 table 历史的示例:
SQL> create table thistory (id,fk_user,state,timestamp)
2 as
3 select 1, 'Me', 'D', sysdate from dual union all
4 select 2, 'Me', 'C', sysdate-1 from dual union all
5 select 3, 'Me', 'B', sysdate-2 from dual union all
6 select 4, 'Me', 'A', sysdate-3 from dual union all
7 select 5, 'You', 'B', sysdate-11 from dual union all
8 select 6, 'You', 'A', sysdate-12 from dual
9 /
Table created.
SQL> exec dbms_stats.gather_table_stats(user,'thistory')
PL/SQL procedure successfully completed.
SQL> alter session set statistics_level = all
2 /
Session altered.
SQL> set serveroutput off
首先是带有子查询的 pre-8i 变体:
SQL> select *
2 from thistory
3 where timestamp =
4 ( select max(timestamp)
5 from thistory latest
6 where fk_user = thistory.fk_user
7 )
8 /
ID FK_USER S TIMESTAMP
---------- ------- - -------------------
1 Me D 19-05-2011 11:20:48
5 You B 08-05-2011 11:20:48
2 rows selected.
SQL> select * from table(dbms_xplan.display_cursor(null,null,'allstats last'))
2 /
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------
SQL_ID 306v8p42zdz34, child number 0
-------------------------------------
select * from thistory where timestamp = ( select max(timestamp) from thistory latest
where fk_user = thistory.fk_user )
Plan hash value: 2894184026
----------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem |
----------------------------------------------------------------------------------------------------------------------
|* 1 | HASH JOIN | | 1 | 2 | 2 |00:00:00.01 | 7 | 1155K| 1155K| 478K (0)|
| 2 | VIEW | VW_SQ_1 | 1 | 2 | 2 |00:00:00.01 | 3 | | | |
| 3 | HASH GROUP BY | | 1 | 2 | 2 |00:00:00.01 | 3 | | | |
| 4 | TABLE ACCESS FULL| THISTORY | 1 | 6 | 6 |00:00:00.01 | 3 | | | |
| 5 | TABLE ACCESS FULL | THISTORY | 1 | 6 | 6 |00:00:00.01 | 4 | | | |
----------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("TIMESTAMP"="VW_COL_1" AND "FK_USER"="THISTORY"."FK_USER")
22 rows selected.
现在解析函数变体:
SQL> select *
2 from ( select h.*
3 , row_number() over (partition by h.fk_user order by timestamp desc) as rn
4 from thistory h
5 )
6 where rn = 1
7 /
ID FK_USER S TIMESTAMP RN
---------- ------- - ------------------- ----------
1 Me D 19-05-2011 11:20:48 1
5 You B 08-05-2011 11:20:48 1
2 rows selected.
SQL> select * from table(dbms_xplan.display_cursor(null,null,'allstats last'))
2 /
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------
SQL_ID b7zscht24wa2s, child number 0
-------------------------------------
select * from ( select h.* , row_number() over (partition by h.fk_user order by timestamp desc)
as rn from thistory h ) where rn = 1
Plan hash value: 2357375523
--------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem |
--------------------------------------------------------------------------------------------------------------------------
|* 1 | VIEW | | 1 | 6 | 2 |00:00:00.01 | 3 | | | |
|* 2 | WINDOW SORT PUSHED RANK| | 1 | 6 | 6 |00:00:00.01 | 3 | 9216 | 9216 | 8192 (0)|
| 3 | TABLE ACCESS FULL | THISTORY | 1 | 6 | 6 |00:00:00.01 | 3 | | | |
--------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN"=1)
2 - filter(ROW_NUMBER() OVER ( PARTITION BY "H"."FK_USER" ORDER BY INTERNAL_FUNCTION("TIMESTAMP") DESC )<=1)
21 rows selected.
最后是使用聚合函数的最有效变体:
SQL> select max(id) keep (dense_rank last order by timestamp) id
2 , fk_user
3 , max(state) keep (dense_rank last order by timestamp) state
4 , max(timestamp)
5 from thistory
6 group by fk_user
7 /
ID FK_USER S MAX(TIMESTAMP)
---------- ------- - -------------------
1 Me D 19-05-2011 11:20:48
5 You B 08-05-2011 11:20:48
2 rows selected.
SQL> select * from table(dbms_xplan.display_cursor(null,null,'allstats last'))
2 /
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------
SQL_ID 9gcbnuf776q27, child number 0
-------------------------------------
select max(id) keep (dense_rank last order by timestamp) id , fk_user , max(state) keep
(dense_rank last order by timestamp) state , max(timestamp) from thistory group by fk_user
Plan hash value: 76026975
--------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem |
--------------------------------------------------------------------------------------------------------------------
| 1 | SORT GROUP BY | | 1 | 2 | 2 |00:00:00.01 | 3 | 9216 | 9216 | 8192 (0)|
| 2 | TABLE ACCESS FULL| THISTORY | 1 | 6 | 6 |00:00:00.01 | 3 | | | |
--------------------------------------------------------------------------------------------------------------------
14 rows selected.
如果您仔细查看计划,您会发现最后一个是最有效的。 而且它不需要内联视图。
希望这会有所帮助。
问候,
抢。
【讨论】: