Oracle：connect by 不带开头，但 group by + 非空唯一列：group by 被忽略

问题描述 投票：0回答：1

我有一个包含三列的表（使用Oracle 19c），其中两列代表父子关系，另一列代表文本描述。我使用不带 

connect by
的 
start with
，但后跟 
group by
，如果描述列是带有 
group by
限制的 
not null
列，Oracle 决定不应用 
unique
，从而导致由于缺少 
start with
子句而重复的行将包含在最终输出中。

create table test_hie (id int, parent int, name varchar2(64) not null);
insert into test_hie (id, parent, name) values (0, null, 'ABC');
insert into test_hie (id, parent, name) values (1, 0, 'DEF');
create unique index test_hie_idx_name on test_hie (name);

alter session set statistics_level = all;

select id from test_hie connect by prior id = parent group by id;
select * from table(dbms_xplan.display_cursor('6pfqf6fg5crck', 0, 'ALLSTATS LAST PROJECTION'));

select id from test_hie connect by prior id = parent group by id, name;
select * from table(dbms_xplan.display_cursor('g56y8n3pubzud', 0, 'ALLSTATS LAST PROJECTION'));

如果没有 

group by
或 
start with
，行 
1
将被包含两次：作为其自己树的根和作为 
0
的子级。

要删除重复项，我应用了 

group by
。在第一个查询中，由于 
str
不包含在 
group by
列表中，行 
1
仅发生一次，但如果我将 
str
包含在 
group by
中，那么它会发生两次。

Output of query 1 (str not in the group by)
id 
---
1
0

Output of query 2 (str in the group by)

id
---
1
0
1

这种行为是不正确的。如果我按 

id
分组，我不应该看到任何重复的 
id
。

将 

str
添加到列列表不会改变任何内容，也不会使用任何聚合函数或向表添加更多列。只需要 
str
成为具有唯一索引的不可空列即可观察不正确的行为。

第一个查询的执行计划；请注意，该计划包含 

HASH GROUP BY
和 
CONNECT BY WITHOUT FILTERING (UNIQUE)
:

SQL_ID  6pfqf6fg5crck, child number 0
-------------------------------------
select id from test_hie connect by prior id = parent group by id
 
Plan hash value: 2961238377
 
----------------------------------------------------------------------------------------------------------------------------------------
| Id  | Operation                              | Name     | Starts | E-Rows | A-Rows |   A-Time   | Buffers |  OMem |  1Mem | Used-Mem |
----------------------------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                       |          |      1 |        |      2 |00:00:00.01 |       7 |       |       |          |
|   1 |  HASH GROUP BY                         |          |      1 |      2 |      2 |00:00:00.01 |       7 |  1968K|  1968K|  647K (0)|
|*  2 |   CONNECT BY WITHOUT FILTERING (UNIQUE)|          |      1 |        |      3 |00:00:00.01 |       7 |  2048 |  2048 | 2048  (0)|
|   3 |    TABLE ACCESS FULL                   | TEST_HIE |      1 |      2 |      2 |00:00:00.01 |       7 |       |       |          |
----------------------------------------------------------------------------------------------------------------------------------------
 
Predicate Information (identified by operation id):
---------------------------------------------------
 
   2 - access("PARENT"=PRIOR NULL)
 
Column Projection Information (identified by operation id):
-----------------------------------------------------------
 
   1 - (rowset=256) "ID"[NUMBER,22]
   2 - "ID"[NUMBER,22], "PARENT"[NUMBER,22], PRIOR NULL[22], LEVEL[4]
   3 - "ID"[NUMBER,22], "PARENT"[NUMBER,22]
 
Note
-----
   - dynamic statistics used: dynamic sampling (level=2)

第二次查询的计划；请注意， 

group by
操作已消失，并且 
CONNECT BY WITHOUT FILTERING
也不再是唯一的：

SQL_ID  g56y8n3pubzud, child number 0
-------------------------------------
select id from test_hie connect by prior id = parent group by id, name
 
Plan hash value: 4109999158
 
------------------------------------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name     | Starts | E-Rows | A-Rows |   A-Time   | Buffers |  OMem |  1Mem | Used-Mem |
------------------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |          |      1 |        |      3 |00:00:00.01 |       7 |       |       |          |
|*  1 |  CONNECT BY WITHOUT FILTERING|          |      1 |        |      3 |00:00:00.01 |       7 |  2048 |  2048 | 2048  (0)|
|   2 |   TABLE ACCESS FULL          | TEST_HIE |      1 |      2 |      2 |00:00:00.01 |       7 |       |       |          |
------------------------------------------------------------------------------------------------------------------------------
 
Predicate Information (identified by operation id):
---------------------------------------------------
 
   1 - access("PARENT"=PRIOR NULL)
 
Column Projection Information (identified by operation id):
-----------------------------------------------------------
 
   1 - "ID"[NUMBER,22], "PARENT"[NUMBER,22], PRIOR NULL[22], LEVEL[4]
   2 - "ID"[NUMBER,22], "PARENT"[NUMBER,22]
 
Note
-----
   - dynamic statistics used: dynamic sampling (level=2)

这是 Oracle 19c 的错误吗？看起来它以某种方式“认为”重复是不可能的（因为非空的唯一列），因此分组是不必要的，忘记了 

connect by
是人为地在行集中注入重复的行。

顺便问一下，

unique
在
CONNECT BY WITHOUT FILTERING (UNIQUE)
中的真正含义是什么？

oracle hierarchy
1个回答
0
投票

是的，这对我来说看起来像是一个错误（我会记录下来）。我怀疑我们对唯一索引的读取“太多”了。我们看到 GROUP BY 'unique col' 并说“酷，我们不需要分组依据”，这可以从 10053 跟踪中看出

QB before group-by removal:******* UNPARSED QUERY IS *******
SELECT "TEST_HIE"."ID" "ID" FROM "SCOTT"."TEST_HIE" "TEST_HIE" CONNECT BY PRIOR "TEST_HIE"."ID"="TEST_HIE"."PARENT" GROUP BY "TEST_HIE"."ID","TEST_HIE"."NAME"
QB before group-by elimination:******* UNPARSED QUERY IS *******
SELECT "TEST_HIE"."ID" "ID" FROM "SCOTT"."TEST_HIE" "TEST_HIE" CONNECT BY PRIOR "TEST_HIE"."ID"="TEST_HIE"."PARENT" GROUP BY "TEST_HIE"."ID","TEST_HIE"."NAME"
Registered qb: SEL$47952E7A 0x78ddb940 (ELIMINATION OF GROUP BY SEL$1; SEL$1)
---------------------
QUERY BLOCK SIGNATURE
---------------------
  signature (): qb_name=SEL$47952E7A nbfros=1 flg=0
    fro(0): flg=0 objn=189661 hint_alias="TEST_HIE"@"SEL$1"

QB after group-by elimination:******* UNPARSED QUERY IS *******
SELECT "TEST_HIE"."ID" "ID" FROM "SCOTT"."TEST_HIE" "TEST_HIE" CONNECT BY PRIOR "TEST_HIE"."ID"="TEST_HIE"."PARENT"
Registered qb: SEL$9BB7A81A 0x78ddb940 (ELIMINATION OF GROUP BY SEL$47952E7A; SEL$47952E7A)

作为临时措施，您可以添加冗余的 START WITH

SQL> create table test_hie (id int, parent int, name varchar2(64) not null);

Table created.

SQL> insert into test_hie (id, parent, name) values (0, null, 'ABC');

1 row created.

SQL> insert into test_hie (id, parent, name) values (1, 0, 'DEF');

1 row created.

SQL> create unique index test_hie_idx_name on test_hie (name);

Index created.

SQL>
SQL> select id from test_hie connect by prior id = parent group by id;

        ID
----------
         1
         0

SQL> select id from test_hie connect by prior id = parent group by id, name;

        ID
----------
         1
         0
         1

SQL> select id from test_hie start with 1=1 connect by prior id = parent group by id, name;

        ID
----------
         1
         0
最新问题
© www.soinside.com 2019 - 2024. All rights reserved.