Thanks for the question regarding "connect by ", versi

Home > Question Details 

Questions Resources Archives Links Popular Hot Files 

gaurang -- Thanks for the question regarding "connect by ", version 8.1.6 

Submitted on 11-Apr-2001 13:05 Central time zone Tom's latest followup | Bookmark | Bottom 

Last updated 30-Mar-2012 7:17 

You Asked 

hi tom 

can u explain me in detail about "start with connect by" sql statement (tree 

structure).i know there is documentatin but it is very confusing. 

and we said... 

It builds a hierarchical query. 

There are 2 components to is: 

"start with" -- this identifies all LEVEL=1 nodes in the tree 

"connect by" -- describes how to walk from the parent nodes above to their children and 

their childrens children. 

Easiest to use an example on emp. If we start with "where mgr is NULL", we generate the 

set of employees that have no mgr (they are the top of the tree). If we 

CONNECT BY PRIOR EMPNO = /* current */ MGR 

that will take all of the PRIOR records (the start with at first) and find all records 

such that the MGR column equals their EMPNO (find all the records of people managed by 

the people we started with). 

Using EMP, the start with SET is: 

scott@ORA8I.WORLD> select ename, empno, mgr from emp 

2 where mgr is null; 

ENAME EMPNO MGR 

---------- ---------- ---------- 

KING 7839 

Now, if we do the "connect by manually" we would find: 

scott@ORA8I.WORLD> select ename, empno, mgr 

2 from emp where mgr = 7839; 


---------- ---------- ---------- 

JONES 7566 7839 

BLAKE 7698 7839 

CLARK 7782 7839 

scott@ORA8I.WORLD> 

KINGS empno is the prior empno. If we build the entire hierarch -- we have: 

scott@ORA8I.WORLD> select lpad(' ',level*2,' ')||ename ename, empno, mgr 

2 from emp 

3 START WITH MGR IS NULL 

4 CONNECT BY PRIOR EMPNO = MGR 

5 / 


--------------- ---------- ----------

KING 7839 

JONES 7566 7839 

SCOTT 7788 7566 

ADAMS 7876 7788 

FORD 7902 7566 

SMITH 7369 7902 

BLAKE 7698 7839 

ALLEN 7499 7698 

WARD 7521 7698 

MARTIN 7654 7698 

TURNER 7844 7698 

JAMES 7900 7698 

CLARK 7782 7839 

MILLER 7934 7782 

14 rows selected. 

So, KING is the start with set then JONES BLAKE and CLARK fall under him. Each of them 

becomes the PRIOR record in turn and their trees are expanded. 

Reviews 

April 12, 2001 - 8am Central time zone Bookmark | Bottom | Top 

Reviewer: gaurang 

useful summary of connect by September 25, 2002 - 2pm Central time zone Bookmark | Bottom | Top 

Reviewer: RParr from Seattle, WA U.S.A. 

in a fraction of the space you provided a much better overview of connect by and basic heirarchical 

query. 

Nice overview January 22, 2003 - 1am Central time zone Bookmark | Bottom | Top 

Reviewer: Anirudh from New delhi, India 

Hi Tom 

Your explaination about the connect by clause was very helpfull. However, I have a question 

we have got a function 

FUNCTION Get_Parent_Entity_Id 

( 

in_test_pgm_id number, 

in_test_admin_id number, 

ic_child_entity_type_code varchar2, 

in_child_entity_id number, 

ic_parent_entity_type_code varchar2 

) 

RETURN number IS 

ln_count NUMBER; 

BEGIN 

ln_count := 0; 

BEGIN 

select n_parent_entity_id 

into ln_count 

from 

( 

select 

n_test_pgm_id, 

n_test_admin_id, 

c_parent_entity_type_code, 

n_parent_entity_id 

from 

rpt_entity_struc res 

start with 

res.n_test_pgm_id = in_test_pgm_id and 

res.n_test_admin_id = in_test_admin_id and 

res.n_entity_struc_id = 0 and 

res.c_child_entity_type_code = ic_child_entity_type_code and 

res.n_child_entity_id = in_child_entity_id 

connect by 

res.n_test_pgm_id = prior n_test_pgm_id and 

res.n_test_admin_id = prior n_test_admin_id and 

res.n_entity_struc_id = prior n_entity_struc_id and

es.c_child_entity_type_code = prior c_parent_entity_type_code and 

res.n_child_entity_id = prior n_parent_entity_id 

) 

where c_parent_entity_type_code = ic_parent_entity_type_code; 

EXCEPTION 

WHEN NO_DATA_FOUND THEN 

ln_count := 0; 

WHEN OTHERS THEN 

ln_err_num := SQLCODE; 

lc_err_msg := 'Error in Get_Parent_Entity_Id - ' || substr(sqlerrm, 1, 200); 

raise_application_error(-20000, lc_err_msg); 

END; 

RETURN ln_count; 

-- 

Algo:n_parent_entity_id is retrieved based on the input parameters passed besides the 

ic_parent_entity_type_code which is one of the important input columns to be considered. The 

Hierarchical query (STARTWITH and CONNECT BY) clauses are used to find the root parent id. 

Question 

The connect by prior clause here refers to the columns and one of those (n_entity_struc_id ) is not 

in the select list. My question is what is the use of that PRIORing the column when it has not been 

selected.? 

Followup January 22, 2003 - 8am Central time zone: 

the same as when you use it in a predicate, no different. 

select ename from emp where empno = :x 

what is the use of empno in the predicate if it has not been selected? well, it is used to 

identify what record(s) you want. same with columns in the start with, connect by prior and so 

on... 

Urgent Requirement May 22, 2003 - 10am Central time zone Bookmark | Bottom | Top 

Reviewer: Shrikant Gavas from India 

Please go throught the script : 

CREATE TABLE MY_LEVEL1 ( 

ORIG_RECP VARCHAR2 (10), 

ORIG_AMT NUMBER, 

REF_RECP VARCHAR2 (10), 

REF_AMT NUMBER, 

PARTY VARCHAR2 (6) ) ; 

INSERT INTO MY_LEVEL1 ( ORIG_RECP, ORIG_AMT, REF_RECP, REF_AMT, 

PARTY ) VALUES ( 

'100', 10000, NULL, 500, 'A0001'); 



'110', 300, '100', 100, 'A0001'); 



'120', 200, '110', 50, 'A0001'); 



'130', 100, '120', 30, 'A0001'); 



'300', 10000, NULL, 500, 'A0001'); 



'310', 300, '300', 100, 'A0001'); 



'320', 200, '310', 50, 'A0001'); 



'330', 100, '320', 30, 'A0001'); 

INSERT INTO MY_LEVEL1 ( ORIG_RECP, ORIG_AMT, REF_RECP, REF_AMT,


'100', 10000, NULL, 500, 'A0003'); 



'110', 300, '100', 100, 'A0003'); 



'120', 200, '110', 50, 'A0003'); 



'130', 100, '120', 30, 'A0003'); 



'100', 10000, NULL, 500, 'A0004'); 



'110', 300, '100', 100, 'A0004'); 



'120', 200, '110', 50, 'A0004'); 



'130', 100, '120', 30, 'A0004'); 



'130', 100, '120', 30, 'A0004'); 

select level x, 

party, orig_recp, ref_recp, orig_amt, ref_amt 

from my_level1 

where party between 'A0001' AND 'A0003' 


prior orig_recp = ref_recp and 

prior party = party 

start with ref_recp is null 

1 A0001 100 10000 500 

2 A0001 110 100 300 100 

3 A0001 120 110 200 50 

4 A0001 130 120 100 30 

1 A0001 300 10000 500 

2 A0001 310 300 300 100 

3 A0001 320 310 200 50 

4 A0001 330 320 100 30 

1 A0003 100 10000 500 

2 A0003 110 100 300 100 

3 A0003 120 110 200 50 

4 A0003 130 120 100 30 

Clients desired output : 

1 A0001 100 10000 500 

4 A0001 130 120 100 30 

1 A0001 300 10000 500 

4 A0001 330 320 100 30 

1 A0003 100 10000 500 

4 A0003 130 120 100 30 

i.e. clients requirement is that they want first and last row per level. We 

have tried a lot but not got any appropriate solution for this. 

Please provide some solution for above query asap. 

Interpret July 13, 2003 - 2pm Central time zone Bookmark | Bottom | Top 

Reviewer: Nitin from Atlanta 

How do I interpret this SQL and the output? Is the output correct? 

SELECT 

TO_CHAR(A.EFFDT,'YYYY-MM-DD'), A.TREE_NODE_NUM, A.TREE_NODE, 

A.TREE_NODE_NUM_END, A.TREE_LEVEL_NUM, A.TREE_NODE_TYPE, A.PARENT_NODE_NUM, A.OLD_TREE_NODE_NUM 

From PSTREENODE A 

Where A.TREE_NAME = 'OVER_EXP' 

And A.SETID = 'LOCKE'

And A.SETID = 'LOCKE' 

And A.EFFDT = (Select Max(Z.EFFDT) 

From PSTREENODE Z 

Where Z.TREE_NAME = A.TREE_NAME 

And Z.TREE_NODE = A.TREE_NODE 

And Z.SETID = A.SETID) 

Start With TREE_NODE = 'WORK' 

Connect By PARENT_NODE_NUM = Prior TREE_NODE_NUM 

TO_CHAR(A. TREE_NODE_NUM TREE_NODE TREE_NODE_NUM_END TREE_LEVEL_NUM T PARENT_NODE_NUM O 

---------- ------------- -------------------- ----------------- -------------- - --------------- - 

2002-01-01 222222223 WORK 333333333 0 G 1 N 

2002-01-01 1222222222 CASE 1333333332 0 G 1111111111 N 

2002-01-01 1166666666 DEPT 1222222221 0 G 1111111111 N 

2002-01-01 1222222222 CASE 1333333332 0 G 1111111111 N 

2002-01-01 1166666666 DEPT 1222222221 0 G 1111111111 N 

Followup July 14, 2003 - 12am Central time zone: 

the output is correct -- given the query. 

However, without any clue as to the "question" - no one can really tell you if the output correctly 

answers your question! 

Clarification July 13, 2003 - 11pm Central time zone Bookmark | Bottom | Top 

Reviewer: Nitin from Atlanta 

The SQL provided in my previous feedback finally made sense. I have the following question. 

The output is not what is desired. We need the condition 

Where A.TREE_NAME = 'OVER_EXP' 

And A.SETID = 'LOCKE' 

And A.EFFDT = (Select Max(Z.EFFDT) 

From PSTREENODE Z 

Where Z.TREE_NAME = A.TREE_NAME 

And Z.TREE_NODE = A.TREE_NODE 

And Z.SETID = A.SETID) 

to be applied prior to executing the CONNECT BY - PRIOR part of the SQL. 

Now, we observe that the SQL is first performing the CONNECT BY - PRIOR and then the WHERE portion. 

Please let us know how it can be achieved. 

Thanks 


then that condition should be in the connect by itself as well. you don't have to only do prior's 

in there. 

start with /connect by is done 

AND THEN 

the where clause is applied 

If you want to stop building the tree when you hit some condition -- put that condition in the 


chained connect by July 26, 2003 - 3am Central time zone Bookmark | Bottom | Top 

Reviewer: farweeda from qatar 

ihave docs tab containing two fields (doc_id , doc_rel_id) 

with e.g. following values {(5,2)(9,5)(11,2)(3,1)(4,2)(8,3)(6,4)} the doc_rel_id has some relation 

with doc_id and vise versa so when i'm asking for the related docs to specific doc_id e.g. doc_id

with doc_id and vise versa so when i'm asking for the related docs to specific doc_id e.g. doc_id 

=2 the resault should be : ( 5,11,4,9 because of 5 ,6 because of 4 ) and if i change doc_id to any 

of the resault's value suppose 4 the same resault should be there 

Followup July 26, 2003 - 12pm Central time zone: 

ops$tkyte@ORA920LAP> select * 

2 from t 

3 start with doc_rel_id = 2 

4 connect by prior doc_id = doc_rel_id; 

DOC_ID DOC_REL_ID 

---------- ---------- 

5 2 

9 5 

11 2 

4 2 

6 4 

To Srikant for his urgent query July 27, 2003 - 1pm Central time zone Bookmark | Bottom | Top 

Reviewer: A reader 

I tried out your scenario - I got a query to get the 

same set of records that you wanted but not in the same 

order - you can probably tweak it. 

Tom may have an even better solution of course! 

Menon:) 

Here goes: 

select x, party, orig_recp, ref_recp, orig_amt, ref_amt 

from 

( 

select a.*, first_value( x) over (partition by party order by party) first_x, 

last_value( x) over (partition by party order by party) last_x 

from 

( 

select level x, party, orig_recp, ref_recp, orig_amt, ref_amt 







) a 

) b 

where x = first_x or 

x = last_x; 

Skipping Gaps? November 23, 2003 - 12pm Central time zone Bookmark | Bottom | Top 

Reviewer: Doug from CT, USA 

Tom - is it possible to make a connect by skip gaps in the sequence? Like 

SQL> desc a; 

Name Null? Type 

----------------------------------------- -------- ---------------------- 

ID NUMBER 

PRINUMBER NUMBER 

SQL> select * from a; 

ID PRINUMBER 

---------- ---------- 

1 2 

2 3 

3 4 

9 10

9 10 

SQL> select id, prinumber from 

2 a start with id=1 

3 connect by prior prinumber=id; 

ID PRINUMBER 

---------- ---------- 

1 2 

2 3 

3 4 

Is there any way to move on to NINE and continue? 

Followup November 23, 2003 - 2pm Central time zone: 

what would be the possible use of this? 

how would this be any different from "select * from t" 

fair point November 23, 2003 - 3pm Central time zone Bookmark | Bottom | Top 


Fair enough - maybe I am asking the wrong question. 

What I'm trying to do is take a time history, potentially with workspace manager that might look 

like this: 

SQL> select name, salary, starttime, stoptime, salary from doug; 

NAME SALARY STARTTIME STOPTIME SALARY 

---------- ---------- --------- --------- ---------- 

Bob 6000 01-JAN-95 01-JUN-95 6000 

Bob 7000 01-JUN-95 01-OCT-95 7000 

Bob 7000 01-OCT-95 01-FEB-96 7000 

Bob 7000 01-FEB-96 01-JAN-97 7000 

Bob 6000 01-JAN-97 01-MAR-98 6000 

Bob 5000 01-APR-03 01-JUN-03 5000 

The startimes and stoptimes connect very nicely except there is a gap between March 1998 and April 

2003. 

If I want to order these dates and include a special salary "gap". We "don't know".. what Bob was 

doing between March,98 and April 2003. 

This is what happens at the root of the query I am looking at - 

SQL> l 

1 select name, salary, starttime, stoptime,decode( lag(salary) over (order by 

starttime), salary, 

2* to_number(null), row_number() over (order by starttime) ) rn from doug 

SQL> / 

NAME SALARY STARTTIME STOPTIME RN 

---------- ---------- --------- --------- ---------- 

Bob 6000 01-JAN-95 01-JUN-95 1 

Bob 7000 01-JUN-95 01-OCT-95 2 

Bob 7000 01-OCT-95 01-FEB-96 

Bob 7000 01-FEB-96 01-JAN-97 

Bob 6000 01-JAN-97 01-MAR-98 5 

Bob 5000 01-APR-03 01-JUN-03 6 


This is WHAT I WANT more or less - the null values when salarys are the same so they can be 

coalesed. What I REALLY want is THIS: 

NAME SALARY STARTTIME STOPTIME RN 

---------- ---------- --------- --------- ---------- 

Bob 6000 01-JAN-95 01-JUN-95 1 

Bob 7000 01-JUN-95 01-OCT-95 2 

Bob 7000 01-OCT-95 01-FEB-96 

Bob 7000 01-FEB-96 01-JAN-97 

Bob 6000 01-JAN-97 01-MAR-98 5

Bob 6000 01-JAN-97 01-MAR-98 5 

Bob NULL 01-MAR-98 01-APR-03 6 

Bob 5000 01-APR-03 01-JUN-03 7 

Now if I just order by time with a "select * from T".. how could I "fill in" the gaps in the 

sequence? 

Thanks, 

D. 

Followup November 23, 2003 - 5pm Central time zone: 

it was the connnect by that really confused me -- couldn't understand where that is coming in. 

Now that I know the question, an answer can be forthcoming :) 

Here is one technique: 

ops$tkyte@ORA920> select name, 

2 decode( r, 1, null, '' ) msg, 

3 decode( r, 1, salary, null ) salary, 

4 decode( r, 1, starttime, last_stop ) starttime, 

5 decode( r, 1, stoptime, starttime ) stoptime 

6 from ( 

7 select name, salary, 

8 starttime, stoptime, 

9 lag(stoptime) over (partition by name order by starttime) last_stop, 

10 decode( lag(stoptime) over (partition by name order by starttime), starttime, null, 

null, null, 1 ) dup_me 

11 from t 

12 ), 

13 (select 1 r from dual union all select 2 r from dual) 

14 where r = 1 or dup_me = 1 

15 order by 4 

16 / 

NAME MSG SALARY STARTTIME STOPTIME 

------------------------------ --------- ---------- --------- --------- 

Bob 6000 01-JAN-95 01-JUN-95 

Bob 7000 01-JUN-95 01-OCT-95 

Bob 7000 01-OCT-95 01-FEB-96 

Bob 7000 01-FEB-96 01-JAN-97 

Bob 6000 01-JAN-97 01-MAR-98 

Bob 01-MAR-98 01-APR-03 

Bob 5000 01-APR-03 01-JUN-03 


Stunning November 23, 2003 - 7pm Central time zone Bookmark | Bottom | Top 


Tom - that is very crafty. I learn a lot about SQL from you. Is there a bit of inefficiency built 

into that however? The join with the 2 rows in the dual table, doesn't that increase the work for 

the db? Don't get me wrong I'm not complaining. :-) Very nice solution. I wasn't sure it could 

be done. Analytic functions seem to have all sorts of interesting uses. 

Followup November 24, 2003 - 7am Central time zone: 

well -- in order to "make up" data we need to join -- to synthesize that row, it was somewhat 

unavoidable (joining to a two row table). pipelined functions and procedural code might have 

worked as well, but this is easier. 

connect by but parent in different table December 2, 2003 - 2am Central time zone 

Bookmark | Bottom | Top 

Reviewer: umesh from blore india 

Tom

Tom 

I have a situation 

There are 2 tables srch_criteria, srch_hierarchy 

srch_criteria is the master 

-----------------------------------------------------id 

name 

07 Aerospace Tier 1 

08 Functional Tier 2 

09 Finance Tier 1 

10 x Tier 1 

11 y Tier 1 

12 z Tier 3 

..... 

srch_hierarchy has the hierarchy maintained in it 

sc_id_child sc_id_parent 

------------------------------------- 

08 07 

09 07 

081 08 

082 08 

091 09 

092 09 

search hierarchy will have a hierachy 

07 is linked with 08,09 

08 has children in 081 , 082 

09 has children in 091, 092 

( I agree bad design) 

I take only Tier 1 from the top and traverse thru whole of children and grand children in the 

hierarchy table 

I need a query for it Is it possible in SQL or should i write procedure or function 

Followup December 2, 2003 - 8am Central time zone: 

I don't see "a bad design" here? looks pretty typical. 

I do not know what "tier 1" from the "top" is though. hows about you show us the desired output 

and explain how it was arrived at. 

using an order by December 2, 2003 - 11am Central time zone Bookmark | Bottom | Top 


Tom - as you pointed out to me earlier, in some situations gaps or duplicates with a connect by are 

redundant because they are the same as an order by clause. 

In a situation where I want to ensure the order but not select the column I am ordering by - will 

this work? 

select T.x, T.y from (select x,y,z from R order by z ) T 

It the order ensured? 

Followup December 2, 2003 - 12pm Central time zone: 

just 

select x, y 

from R 

order by Z;

December 9, 2003 - 2am Central time zone Bookmark | Bottom | Top 

Reviewer: umesh from blore 

SQL> l 

1 SELECT 

2 DECODE ( ll ,1 , crit1.name ) one , 

3 DECODE ( ll ,2 , crit1.name) two , 

4 DECODE ( ll ,3 , crit1.name) three , 

5 DECODE ( ll ,4 , crit1.name) four 

6 FROM 

7 (SELECT sc_id_child , sc_id_parent , LEVEL ll FROM CNT_SEARCH_HIERARCHY src 

8 START WITH sc_id_parent='crite000000000000002' 

9 CONNECT BY PRIOR sc_id_child=sc_id_parent 

10 ) , CNT_SEARCH_CRITERIA crit , CNT_SEARCH_CRITERIA crit1 

11 WHERE crit.id = sc_id_parent 

12* AND crit1.id=sc_id_child 

SQL> / 

ONE TWO THREE FOUR 

-------------------------- --------------------------- ------------------------ 

------------------- 

Building Automation 

ACSELON 

Authorized Trainer 

Programs 

Delivery & Installation 

ADEPT 

Project 

Management 

Human Resources 

Love it, but ... March 4, 2004 - 3pm Central time zone Bookmark | Bottom | Top 

Reviewer: Peter Tran from Houston, TX USA 

Hi Tom, 

What happens if you have a dataset where the top mgr refers to himself? Using your example, if the 

MGR = 7839 for KING and I run the same query you have Oracle gives me a: 

ORA-01436: CONNECT BY loop in user data 

A connect by record cannot reference itself otherwise we have an infinite loop situation. 

Is there anyway I can filter this out? In other words, I still want the answer you got earlier, 

but now I have the situation where KING refers to himself in the MGR column. 


-Peter 

Followup March 4, 2004 - 3pm Central time zone: 

that case is easy -- since we have the "loop" in a single row, we can simply filter it out in the 

"connect by" clause: 

ops$tkyte@ORA9IR2> create table emp as select * from scott.emp; 

Table created. 

ops$tkyte@ORA9IR2> update emp set mgr = empno where mgr is null; 

1 row updated. 

ops$tkyte@ORA9IR2> commit; 

Commit complete.

ops$tkyte@ORA9IR2> 

ops$tkyte@ORA9IR2> select lpad(' ',level*2,' ')||ename ename, empno, mgr 

2 from emp 

3 START WITH ename = 'KING' 


5 / 

ERROR: 

ORA-01436: CONNECT BY loop in user data 

no rows selected 

ops$tkyte@ORA9IR2> select lpad(' ',level*2,' ')||ename ename, empno, mgr 

2 from emp 


4 CONNECT BY PRIOR EMPNO = MGR AND empno mgr 

5 / 


--------------- ---------- ---------- 

KING 7839 7839 

JONES 7566 7839 

SCOTT 7788 7566 

ADAMS 7876 7788 

FORD 7902 7566 

SMITH 7369 7902 

BLAKE 7698 7839 

ALLEN 7499 7698 

WARD 7521 7698 

MARTIN 7654 7698 

TURNER 7844 7698 

JAMES 7900 7698 

CLARK 7782 7839 

MILLER 7934 7782 


And in 10g, you have NOCYCLE to avoid the loops anywhere: 

ops$tkyte@ORA10G> select lpad(' ',level*2,' ')||ename ename, empno, mgr 

2 from emp 


4 CONNECT BY NOCYCLE PRIOR EMPNO = MGR 

5 / 


--------------- ---------- ---------- 

KING 7839 7839 

JONES 7566 7839 

FORD 7902 7566 

SMITH 7369 7902 

SCOTT 7788 7566 

ADAMS 7876 7788 

BLAKE 7698 7839 

ALLEN 7499 7698 

WARD 7521 7698 

MARTIN 7654 7698 

TURNER 7844 7698 

JAMES 7900 7698 

CLARK 7782 7839 

MILLER 7934 7782 


Awesome... March 4, 2004 - 4pm Central time zone Bookmark | Bottom | Top 


Thanks for the quick turn-around. 

The combination of Tom Kyte and Oracle really rocks. 

Awesome...awesome...awesome.

-Peter 

Can I get the parent position w.r.t. to the rownum? March 5, 2004 - 4pm Central time zone 



Hi Tom, 

Is it possible for me to generate the ParentPosition index w.r.t. to the assigned ROWNUM? Of 

course, ROWNUM starts at 1, but my example below is using base 0. Either way, you can see where 

I'm getting at with the example below. 

Currently, I'm doing this mapping in code. It would be much better if I can do this within SQL. 


-Peter 

ROWNUM ENAME EMPNO MGR ParentPos 

------ --------------- ---------- ---------- ------------- 

0 KING 7839 7839 0 

1 JONES 7566 7839 0 

2 FORD 7902 7566 1 

3 SMITH 7369 7902 2 

4 SCOTT 7788 7566 1 

5 ADAMS 7876 7788 4 

6 BLAKE 7698 7839 0 

7 ALLEN 7499 7698 6 

8 WARD 7521 7698 6 

9 MARTIN 7654 7698 6 

10 TURNER 7844 7698 6 

11 JAMES 7900 7698 6 

12 CLARK 7782 7839 0 

13 MILLER 7934 7782 12 


what is "parentPos" 

ParentPos March 6, 2004 - 5pm Central time zone Bookmark | Bottom | Top 


ParentPos is the RowNum value of the parent. 

King is the parent of Jones, Blake, and Clark which is why ParentPos for them is 0. 

Blake is the parent of Alan, Ward, Martin, Turner, and James, so you see their ParentPos is 6 

because Blake's RowNum is 6. 


-Peter 


scott@ORA9IR2> select rnum, ename, empno, mgr, 

2 substr( scbp2, 1, instr(scbp2,',')-1 ) parentpos 

3 from ( 

4 select a.*, 

5 substr( scbp, instr(scbp, ',', -1, 2 )+1 ) scbp2 

6 from ( 

7 select rownum-1 rnum, rpad( ' ', 2*level, ' ' ) || ename ename, empno, mgr, 

8 sys_connect_by_path( rownum-1, ',' ) scbp 

9 from emp 

10 start with mgr is null 

11 connect by prior empno = mgr 

12 ) a

12 ) a 

13 ) 

14 / 

RNUM ENAME EMPNO MGR PARENT 

---------- -------------------- ---------- ---------- ------ 

0 KING 7839 

1 JONES 7566 7839 0 

2 SCOTT 7788 7566 1 

3 ADAMS 7876 7788 2 

4 FORD 7902 7566 1 

5 SMITH 7369 7902 4 

6 BLAKE 7698 7839 0 

7 ALLEN 7499 7698 6 

8 WARD 7521 7698 6 

9 MARTIN 7654 7698 6 

10 TURNER 7844 7698 6 

11 JAMES 7900 7698 6 

12 CLARK 7782 7839 0 

13 MILLER 7934 7782 12 


Just amazing... March 6, 2004 - 9pm Central time zone Bookmark | Bottom | Top 


Excellent! 

Thanks! 

-Peter 

CONNECT BY March 9, 2004 - 12am Central time zone Bookmark | Bottom | Top 

Reviewer: ANOOP GUPTA from INDIA 

it would very useful for me to understand connect by 

specially to get first and last in tree. 

one solution i also have for geting first and last 

in tree. 

select * 

from 

( 

select level x, party, orig_recp, ref_recp, orig_amt, ref_amt 







) a 

where a.ref_recp is null 

or a.orig_recp not in (select nvl(ref_recp,'NULL') 







) 

its output is similar as your client want 

1 A0001 100 10000 500 

4 A0001 130 120 100 30 

1 A0001 300 10000 500 

4 A0001 330 320 100 30 

1 A0003 100 10000 500 

4 A0003 130 120 100 30

Find Employee's top most manager August 19, 2004 - 6pm Central time zone 


Reviewer: Mita from NJ 

For the following data, 

Emp_id Mgr_id 

1 2 

2 3 

3 4 

4 

5 6 

6 7 

7 

I need the following result 

EmpId TopManager 

1 4 

2 4 

3 4 

4 4 

5 7 

6 7 

7 7 

How can I achieve this ?? 

Followup August 19, 2004 - 8pm Central time zone: 

if you gave me a create table and insert into's, i'd show you how to use a scalar subquery with 

connect by using max to do this. 

maybe that is enough to get you going.... 

Sample Data August 24, 2004 - 12pm Central time zone Bookmark | Bottom | Top 

Reviewer: Mita from NJ 

create table emp (emp_Id number, mgr_Id number); 

insert into emp values(1,2); 



insert into emp values(4,Null); 



insert into emp values(7,Null); 


ops$tkyte@ORA9IR2> select emp_id, 

2 to_number( substr( 

3 (select max( to_char(level,'fm000009') || ' ' || emp_id ) 

4 from emp e2 

5 start with e2.emp_id = e1.emp_id 

6 connect by prior mgr_id is not null and prior mgr_id = emp_id ) 

7 , 8 ) ) top_mgr 

8 from emp e1 

9 / 

EMP_ID TOP_MGR 

---------- ---------- 

1 4 

2 4

2 4 

3 4 

4 4 

5 7 

6 7 

7 7 


More Connect by October 14, 2004 - 5pm Central time zone Bookmark | Bottom | Top 

Reviewer: Vinnie from Orlando 

Tom, 

I have the following: 

create table parent( rowid number(2) primary key, id varchar2(14) ); 

create table child( rowid_parent number(2), id varchar2(14) ); 

insert into parent(1, '12345'); 




insert into child (1,'12346'); 



I would like to pass in a id (i.e. 12345) 

and find all his child (1,'12346') 

Then for each child find his children (1,'12347') and so on down the tree. 

Like the following: 

12345 

12346 

12347 

12348 

Followup October 14, 2004 - 7pm Central time zone: 

what have you tried so far....... 

wacky structure to store parent/child in don't you think? probably could find a less efficient 

structure for a hierarchy.... 

October 15, 2004 - 8am Central time zone Bookmark | Bottom | Top 


create table parent( rowid number(2) primary key, id varchar2(14) ); 

create table child( rowid_parent number(2), id varchar2(14) ); 








I have tried the following: 

select a.id from parent a, child b 

start with a.id='12345' 

connect by a.rowid = b.rowid_parent; 

Was hoping this could be accomplised using this type of approach....somehow!!

Followup October 15, 2004 - 11am Central time zone: 

do you have to live with this "structure"? ugh. it hurts my head to look at it. the names don't 

even make sense. 

Ugh October 15, 2004 - 2pm Central time zone Bookmark | Bottom | Top 


Ugh is right, I have to live with the structure! 

Perhaps I can explain this better. 

I have table EMP with a list of emplyees, and table SUB with each employees subordinates. What I 

need is a list of all the subordinates for a given EMP rolled up to include all subordinate 

emplyees. So if I select employee SMITH I get: 

SMITH 

ADAMS 

JONES 

ARMSTRONG 

FRANK 

JAMES 

BROWN 

But The structure is still the same as described before. 

CREATE TABLE EMP (row_id number, ename varchar2(30)); 

CREATE TABLE SUB (row_id_parent number, ename varchar2(30)); 

Just assume the names are unique for this test case. 

INSERT INTO EMP (1,'SMITH'); 

INSERT INTO EMP (2,'ADAMS'); 

INSERT INTO EMP (3,'JONES'); 

INSERT INTO EMP (4,'ARMSTRONG'); 

INSERT INTO EMP (5,'FRANK'); 

INSERT INTO EMP (6,'JAMES'); 

INSERT INTO EMP (7,'BROWN'); 

INSERT INTO SUB (1,'ADAMS'); 

INSERT INTO SUB (1,'BROWN'); 

INSERT INTO SUB (2,'JONES'); 

INSERT INTO SUB (2,'ARMSTONG'); 

INSERT INTO SUB (2,'JAMES'); 

INSERT INTO SUB (4,'FRANK'); 

Hope this helps with your head ache:) 


ops$tkyte@ORA9IR2> select rpad('*',2*level,'*') || ename name 

2 from (select row_id_parent, ename, (select row_id from emp where ename = sub.ename) row_id 

from sub) 

3 start with row_id_parent = ( select row_id from emp where ename = 'SMITH' ) 

4 connect by prior row_id = row_id_parent 

5 / 

NAME 

------------------------------ 

**ADAMS 

****JONES 

****ARMSTRONG 

******FRANK 

****JAMES 

**BROWN 


I figure "smith" can be "implied", you could union him in, but you already sort of know "smith"

GREAT October 18, 2004 - 1pm Central time zone Bookmark | Bottom | Top 

Reviewer: Vinnie from ORlando 

This works great!! 

Can you explain this all in plain text? 


select row_id_parent, ename, 

(select row_id from emp where ename = sub.ename) row_id 

from sub 

gives us the single table with the parent/child info we need. we needed the row_id_parent and 

row_id together, then connect by is trivial. 

Able to show only certain hierarchies? October 21, 2004 - 12pm Central time zone 


Reviewer: Jon from CT, USA 

Is it possible to only show certain hierarchies that match a WHERE condition? For example in the 

data below, FRENCH reports to both DAVIS AND BLAKE. I would like to show ONLY those hierarchies. 

drop table t1; 

create table t1 (EMP varchar2(30), MGR varchar2(30)); 

insert into t1 values ('ADAMS',null); 

insert into t1 values ('BLAKE','ADAMS'); 

insert into t1 values ('CHARLES','ADAMS'); 

insert into t1 values ('DAVIS','ADAMS'); 

insert into t1 values ('EDWARDS','BLAKE'); 

insert into t1 values ('FRENCH','BLAKE'); 

insert into t1 values ('GAVIN','BLAKE'); 

insert into t1 values ('HOWARD','CHARLES'); 

insert into t1 values ('INGRAHAM','CHARLES'); 

insert into t1 values ('JONES','CHARLES'); 

insert into t1 values ('KING','DAVIS'); 

insert into t1 values ('LEWIS','DAVIS'); 

-- FRENCH REPORTS TO BOTH DAVIS AND BLAKE 

insert into t1 values ('FRENCH','DAVIS'); 

insert into t1 values ('MATTHEWS','FRENCH'); 

insert into t1 values ('NEWMAN','FRENCH'); 

COMMIT; 

SELECT substr(LPAD(' ',2*(LEVEL - 1))||EMP,1,40) Employee 

from t1 

connect by prior 

emp = mgr 

start with mgr is null; 

This results in 

EMPLOYEE 

---------------------------------------- 

ADAMS 

BLAKE 

EDWARDS 

FRENCH 

MATTHEWS 

NEWMAN 

GAVIN 

CHARLES 

HOWARD 

INGRAHAM 

JONES 

DAVIS 

KING 

LEWIS 

FRENCH 

MATTHEWS 

NEWMAN

NEWMAN 

I would like it to only show: 

EMPLOYEE 

---------------------------------------- 

ADAMS 

BLAKE 

EDWARDS 

FRENCH 

MATTHEWS 

NEWMAN 

GAVIN 

DAVIS 

KING 

LEWIS 

FRENCH 

MATTHEWS 

NEWMAN 

Is that possible 


is it possible for there to be multiple "roots" in this? or will french always roll up to a single 

root node? 

Thanks for responding October 21, 2004 - 3pm Central time zone Bookmark | Bottom | Top 


There could be multiple roots. How would the answer differ if the answer was a single root? I ask 

because it may be possible to create a view that makes the multiple roots all point to a single 

(new) root, if that would make this more doable. 


if there were one root, we could "start with" using this: 

ops$tkyte@ORA9IR2> select level, sys_connect_by_path( emp, '/' ) scbp 

2 from t1 

3 start with emp = 'FRENCH' 

4 connect by prior mgr = emp 

5 / 

LEVEL SCBP 

---------- ------------------------- 

1 /FRENCH 

2 /FRENCH/BLAKE 

3 /FRENCH/BLAKE/ADAMS 

1 /FRENCH 

2 /FRENCH/DAVIS 

3 /FRENCH/DAVIS/ADAMS 


ops$tkyte@ORA9IR2> select max( to_char(level,'fm00009') || ' ' || sys_connect_by_path( emp, '/' ) ) 

scbp 

2 from t1 



5 / 

SCBP 

------------------------- 

00003 /FRENCH/DAVIS/ADAMS 

see how we could get ADAMS... but if there are multiple roots that each take a different number 

of levels to get to -- that would be a problem

with one root, you would START WITH EMP = ( SELECT that root ) 

but in hindsight -- i see that would not work either. We'd have to actually run a connect by query 

per row -- just to see if french was in the hierarchy up or down the tree. it'd be very expensive. 

I'd probably rather run two queries and may union all them together -- one that runs "up" the tree 

from french, another that runs "down the tree" from french. 

Further explanation October 21, 2004 - 3pm Central time zone Bookmark | Bottom | Top 


The following demostrates what I meant by creating a view to point multiple roots to a new root: 

drop table t1; 

create table t1 (EMP varchar2(30), MGR varchar2(30)); 

insert into t1 values ('ADAMS',null); 

insert into t1 values ('BLAKE','ADAMS'); 

insert into t1 values ('CHARLES','ADAMS'); 

insert into t1 values ('DAVIS','ADAMS'); 

insert into t1 values ('EDWARDS','BLAKE'); 

insert into t1 values ('FRENCH','BLAKE'); 

insert into t1 values ('GAVIN','BLAKE'); 

insert into t1 values ('HOWARD','CHARLES'); 

insert into t1 values ('INGRAHAM','CHARLES'); 

insert into t1 values ('JONES','CHARLES'); 

insert into t1 values ('KING','DAVIS'); 

insert into t1 values ('LEWIS','DAVIS'); 

-- FRENCH REPORTS TO BOTH DAVIS AND BLAKE 

insert into t1 values ('FRENCH','DAVIS'); 

insert into t1 values ('MATTHEWS','FRENCH'); 

insert into t1 values ('NEWMAN','FRENCH'); 

insert into t1 values('OLIVER',NULL); 

-- FRENCH ALSO REPORTS TO OLIVER 

insert into t1 values('FRENCH','OLIVER'); 

drop view view_t1; 

create view view_t1 as 

select '.' EMP,null MGR from dual 

union 

select emp, nvl(mgr, '.') MGR from t1; 


from t1 


emp = mgr 



from view_t1 


emp = mgr 


To Jon ... October 22, 2004 - 4pm Central time zone Bookmark | Bottom | Top 

Reviewer: Gabe 

So ... how should your report look if FRENCH who now reports to DAVIS gets to lead a new project 

having DAVIS as a resource (it frequently happens in real life)? 

Maybe hierarchical queries are not quite applicable to your _model_. They work on hierarchies ... 

don't work very well on graphs. 


(in 10g with NOCYCLE and isleaf and other new functions -- they will work much much better with 

graphs)

Not quite there yet October 25, 2004 - 9am Central time zone Bookmark | Bottom | Top 

Reviewer: Jon from Jon, CT, USA 

It may be intuitively obvious to you, but I'm struggling with a query that will give me the results 

I need using your "two query - one up, one down - union all" suggestion. I need the query to 

return the following results: 

Level Emp 

1 ADAMS 

2 BLAKE 

3 FRENCH 

4 MATTHEWS 

4 NEWMAN 

2 DAVIS 

3 FRENCH 

4 MATTHEWS 

4 NEWMAN 

1 OLIVER 

2 FRENCH 

3 MATTHEWS 

3 NEWMAN 

Also, although we're not at 10g yet, if that has a more straightforward solution, I'd be interested 

in seeing the solution. 


you'll get the output of two queries -- unioned together. It will not be exactly like that above 

-- it'll be the data you need however -- 

ops$tkyte@ORA9IR2> select level, rpad('*',2*level,'*') || emp ename 

2 from t1 



5 / 

LEVEL ENAME 

---------- ------------------------------ 

1 **FRENCH 

2 ****BLAKE 

3 ******ADAMS 

1 **FRENCH 

2 ****DAVIS 

3 ******ADAMS 

1 **FRENCH 

2 ****OLIVER 


ops$tkyte@ORA9IR2> 

ops$tkyte@ORA9IR2> select level, rpad('*',2*level,'*') || emp ename 

2 from t1 

3 start with mgr = 'FRENCH' 

4 connect by prior emp = mgr 

5 / 

LEVEL ENAME 

---------- ------------------------------ 

1 **MATTHEWS 

1 **NEWMAN 

that is what you'll have to work with if you want just "french" 

November 22, 2004 - 4am Central time zone Bookmark | Bottom | Top 


How to skip the self reference November 29, 2004 - 5am Central time zone Bookmark | Bottom | Top 

Reviewer: Ara

ORACLE : 8i 

Problem: Connect-by because self-reference 

Hi Tom, 

Is there any way to skip the self-reference. 

I need all the records before self-reference. Suppose my table has 1000000 records, there is a 

self-referece at 50000th records. why should my query fail? 

is there any way to skip the this records and move further. 


Followup November 29, 2004 - 8am Central time zone: 

in 10g there is a "NOCYCLE" operation. to detect a connect by loop prior to that was not part of 

the functionality of connect by. 

connect by in view December 1, 2004 - 10am Central time zone Bookmark | Bottom | Top 


i am refering toms comment: 

"see how we could get ADAMS... but if there are multiple roots that each take a 

different number of levels to get to -- that would be a problem" 

we are exactky in that situation, we wannt a connect by, and have a root-node column returned 

select :x, ... 

[...] 

start with :x 

would do the job, but we need the functionality in a view, so we can't use the start with clause. 

one solution would be to 

select substr(sys_connect_by_path, 1, 14) RNODE, ... 

(the column is char(14)) 

the problem is there is a bug with sys_connect_by_path when used in a query. 

select RNODE friom viewxyz; works but using RNODE in a where clause raises an ORA-600 

the problem is, we currently can't update. 

maybe someone has a different solution? 

Regarding Mita's request (08/24/2004) February 11, 2005 - 4am Central time zone 


Reviewer: Jet-Lagged Jim from Vancouver 

Hi Tom. 

So if we wanted a third column in the result set that corresponds to the hierarchical level of each 

manager's employees, how would that be done? Tried several things, but can't quite seem to "get 

it". 

example output: 

emp_id top_mgr hier_level 

------- -------- ----------- 

1 4 4 

2 4 3 

3 4 2 

4 4 1 

5 7 3 

6 7 2 

7 7 1 

Thanks so much!

From Jet-Lagged Jim February 11, 2005 - 1pm Central time zone Bookmark | Bottom | Top 

Reviewer: Jim from Vancouver 

Regading my above inquiry - please disregard, I figured it out. T'was a muddled-brain posting at 

1:00am from a handful of timezones. Feel free to remove. Thanks again. 

Followup February 12, 2005 - 8am Central time zone: 

don't sweat it, i just got back last night 6 hours off myself :) 

2 Tables in using start with , connect by May 9, 2005 - 3pm Central time zone 


Reviewer: Lamya from Houston,TX 

I have 2 tables 

CREATE TABLE HYBRIDOMA 

( 

HYBRIDOMA_ID NUMBER(10), 

HYBRIDOMA_NAME VARCHAR2(20 BYTE), 

) 

CREATE TABLE CLONE 

( 

CLONE_ID NUMBER(10), 

CLONE_NAME VARCHAR2(60 BYTE), 

PARENT_TYPE VARCHAR2(10 BYTE), 

PARENT_ID NUMBER(22) 

) 

INSERT INTO CLONE ( CLONE_ID, CLONE_NAME, PARENT_TYPE, PARENT_ID ) VALUES ( 

1, 'clone_from_h1', 'HYBRID', 1); 


2, 'clone_from_clone1', 'CLONE', 1); 


3, 'clone_from_h1', 'HYBRID', 1); 

commit; 

INSERT INTO HYBRIDOMA ( HYBRIDOMA_ID, HYBRIDOMA_NAME ) VALUES ( 

0, 'fff'); 


1, 'rrrrr'); 


2, 'ddddddd'); 

commit; 

Now the clone table has parent type = clone or hybrid. thus the clone can have parents in the 

hybridoma table. 

I would like to create a select statement which would start on the top and select all children , 

from both hybridoma and clone . 

I tried this but its not helping me . 

select level , lpad('*',level*2,'*') || decode( parent_type ,'CLONE' , clone_name , 'HYBRID' , 

hybridoma_name) 

from clone c , hybridoma h 

where c.parent_id = h.HYBRIDOMA_ID 

start with c.parent_id = 1 

connect by prior c.clone_id = c.parent_id and clone_ID parent_id 

zone 

Very Informative - but can you get only a portion of the hierarchy? January 30, 2006 - 7pm Central time 

Reviewer: Marshall B Thompson from Charlotte, NC, USA 

Bookmark | Bottom | Top

Reviewer: Marshall B Thompson from Charlotte, NC, USA 

late to the party, I know, but I find this thread very informative. Back to the very original 

example you used at the top of the thread, you produced the results: 


--------------- ---------- ---------- 

KING 7839 

JONES 7566 7839 

SCOTT 7788 7566 

ADAMS 7876 7788 

FORD 7902 7566 

SMITH 7369 7902 

BLAKE 7698 7839 

ALLEN 7499 7698 

WARD 7521 7698 

MARTIN 7654 7698 

TURNER 7844 7698 

JAMES 7900 7698 

CLARK 7782 7839 

MILLER 7934 7782 

What if you had the employee numbers for allen, ward, martin, turner, james, and miller, and wanted 

to get the results below. (The hierarchy above just those employees.) How would that be done? 


--------------- ---------- ---------- 

KING 7839 

BLAKE 7698 7839 

ALLEN 7499 7698 

WARD 7521 7698 

MARTIN 7654 7698 

TURNER 7844 7698 

JAMES 7900 7698 

CLARK 7782 7839 

MILLER 7934 7782 


and what if martin was not in the list? 

I must be missing something January 31, 2006 - 8am Central time zone Bookmark | Bottom | Top 

Reviewer: Marshall B Thompson from Charlotte, NC USA 

From your response, this must be obvious, but is not to me at the moment. (But, I would expecte 

Martin to not be in the output, but all else the same.) 

Against my HR sample schema, running the following query: 

select lpad(' ',level*2,' ')||last_name ename, employee_id, manager_id 

from hr.employees 

start with manager_id is null 

connect by prior employee_id = manager_id 

I get: 

King 

100 

Kochhar 

101 100 

Greenberg 

108 101 

Faviet 

109 108 

Chen 

110 108 

Sciarra 

111 108

111 108 

Urman 

112 108 

Popp 

113 108 

Whalen 

200 101 

Mavris 

203 101 

................etc. 

Given employee id's 111 (Sciarra) and 112 (Urman), I'd like to get the relevant hierarchy from them 

up. Desired results would be: 

King 

100 

Kochhar 

101 100 

Greenberg 

108 101 

Sciarra 

111 108 

Urman 

112 108 

But, when I do this: 



where employee_id in (111, 112) 

start with manager_id is null 

connect by prior employee_id = manager_id 

I wind up with: 

Sciarra 

111 108 

Urman 

112 108 

How can I accomplish the desired results? 

Ahhhhhh January 31, 2006 - 10am Central time zone Bookmark | Bottom | Top 

Reviewer: Marshall B Thompson from Charlotte, NC USA 

found my inspiration here: 

http://asktom.oracle.com/pls/asktom/f?p=100:11:::::P11_QUESTION_ID:13129005201417 



start with employee_id in (111, 112) 

connect by prior manager_id = employee_id 

Sciarra 

111 108 

Greenberg 

108 101 

Kochhar 

101 100 

King 

100 

Urman 

112 108

Greenberg 

108 101 

Kochhar 

101 100 

King 

100 

Followup January 31, 2006 - 3pm Central time zone: 

I did not mean to make you think this was "obvious", just that the answer would have to be very 

different if "martin" (or indeed any of the leaves) were not to be in the output! 

Connect by prior February 28, 2006 - 9am Central time zone Bookmark | Bottom | Top 

Reviewer: Sanjeev Vibuthi from Hyderabad, India 

Hi Tom, 

This is my table... I have set all reports_to correctly.. But the rows are repeating when I 

executed the following query... 

desc test_unit 

Name Null? Type 

----------------------------------------------------------------- -------- ------------------------ 

D_CD NOT NULL NUMBER(3) 

U_CD NOT NULL NUMBER(3) 

U_NAME VARCHAR2(50) 

U_TYPE VARCHAR2(10) 

REPORTS_TO NUMBER(3) 

SELECT SUBSTR(lpad(' ',Level*2,' '),1,30)||U_name Name, 

U_Cd, Repors_To 

From test_unit 

where d_cd=32 

start with d_cd=32 and u_cd=500 

connect by prior u_cd = repors_to 

--d_cd=32 where 32 is Top Level 

NAME U_CD REPORS_TO 

-------------------------------------------------------------------------------- ---------- 

------------ 

Central - NY 500 0 

EZ 301* 

500 

EZ-SDivision 202 

301* 

Sdivision2 13 

202 

Sdivision3 70 

202 

EZ-KDivision 203 

301* 

KDivision2 56 

203 

KDivision3 59 

203 

EZ-MDivision 223 

301* 

MDivision3 68 

223 

MDivision2 44 

223 

-- It is repeating (already displayed in "ES_SDivision") 

Sdivision2 13 

202? 

Sdivision3 70 

202? 

-- The following UNITS should come under "SZ" - "SZ-CR Division" 

CR Division-4 37 

221*** 


221

221 

221 


NZ 302 

500 

NZ-GP-Division 211 

302 

GP-Division-3 40 

211 

GP-Division-4 77 

211 

NZ-MK-Division 212 

302 

MK-Division-2 46 

212 


212 


212 

NZ-BP-Division 213 

302 

BP-Division-4 84 

213 

BP-Division-5 91 

213 

-----The followings UNITS already displayed under "EZ" - "EZ-KDivision " 

KDivision2 56 

203 

KDivision3 59 

203 

-----The followings UNITS already displayed under "SZ" - "SZ-CK Division " 

CK Division-3 71 

224 


224 

-- The following Two Zones are Coming correctly 

SZ 303 

500 

SZ-CR Division 221*** 

303 


221 


221 


221 

SZ-MR Division 222 

303 

MR Division-4 86 

222 

MR Division-3 62 

222 

SZ-CK Division 224 

303 


224 


224 

SZ-SR Division 237 

303 

SR Division-4 85 

237 

SR Division-3 64 

237 

WZ 304 

500 

WZ - AN Division 231 

304 

AN Division-4 41 

231 

AN Division-5 74 

231 

WZ - GL Division 232 

304 

GL Division-3 45 

232

GL Division-4 69 

232 

WZ - PG Division 234 

304 

PG Division-1 57 

234 

PG Division-2 67 

234 

WZ - BH Division 236 

304 

BH Division-2 87 

236 

BH Division-1 11 

236 

Where is the error in my question... 

Thanks in Adv.. 

Sanjeev Vibuthi 


I don't know. 

Why don't I know? 

because I don't have anything to reproduce with, no create table, no insert intos, we have no clue 

what you original source data looks like at all. 

So, who knows. 

(can you tell this gets frustrating when day after day no one reads the thing that says 

.... If your followup requires a response that might include a query, you had better supply very 

very simple create tables and insert statements. I cannot create a table and populate it for each 

and every question. The SMALLEST create table possible (no tablespaces, no schema names, just like 

I do in my examples for you) ...) 

February 28, 2006 - 10am Central time zone Bookmark | Bottom | Top 

Reviewer: Alex 

Especially since you have to check a box saying you agreed. You could probably change that to say 

anything and people would just click it. "I really like Michael Bolton he's fantastic...." ;) 

Connect by March 1, 2006 - 3am Central time zone Bookmark | Bottom | Top 

Reviewer: Sanjeev Vibuthi from Hyderabad, India 

Hi Tom, 

I will explain my problem with EMP Table ... 

I added COMPANY_CD column to EMP table and renamed it to TEST_EMP (PK(COMPANY_CD, EMPNO)) 

Test_Emp table contains company wise Employee Details 

SCOTT@ testdb 01-MAR-06>SELECT COMPANY_CD, EMPNO,ENAME,MGR FROM TEST_EMP 

2* ORDER BY COMPANY_CD 

SCOTT@ testdb 01-MAR-06>/ 

COMPANY_CD EMPNO ENAME MGR 

---------- ---------- ---------- ---------- 

101 7369 SMITH 7902 

7499 ALLEN 7698 

7521 WARD 7698 

7566 JONES 7839 

7654 MARTIN 7698

7698 BLAKE 7839 

7782 CLARK 7839 

7788 SCOTT 7566 

7839 KING 

7844 TURNER 7698 

7876 ADAMS 7788 

7900 JAMES 7698 

7902 FORD 7566 

7934 MILLER 7782 

201 7369 SMITH 7902 

7499 ALLEN 7698 

7521 WARD 7698 

7566 JONES 7839 

7654 MARTIN 7698 

7698 BLAKE 7839 

7782 CLARK 7839 

7788 SCOTT 7566 

7839 KING 

7844 TURNER 7698 

7876 ADAMS 7788 

7900 JAMES 7698 

7902 FORD 7566 

7934 MILLER 7782 


-- I have written connect by query to get Employee Hierarchy in a Given Company 

SCOTT@ testdb 01-MAR-06> 

1 Select substr(lpad('-',Level*2,'-')||Empno,1,15) Empno, 

2 ename From test_emp 

3 where company_cd=101 

4 start with company_cd=101 

5* connect by prior empno=mgr 


EMPNO ENAME 

--------------- ---------- 

--7788 SCOTT 

----7876 ADAMS 

--7902 FORD 

----7369 SMITH 

--7499 ALLEN 

--7521 WARD 

--7654 MARTIN 

--7844 TURNER 

--7900 JAMES 

--7934 MILLER 

--7876 ADAMS 

--7566 JONES 

----7788 SCOTT 

------7876 ADAMS 

----7902 FORD 

------7369 SMITH 

------7876 ADAMS 

------7369 SMITH 

--7698 BLAKE 

----7499 ALLEN 

----7521 WARD 

----7654 MARTIN 

----7844 TURNER 

....... 

....... 

....... 

55 rows selected. -- But table is having only 24 records 



2 ename From (Select * from test_emp where company_cd=101) test_emp -- Inline View 

3 start with company_cd=101 



EMPNO ENAME

--------------- ---------- 

--7788 SCOTT 

----7876 ADAMS 

--7902 FORD 

----7369 SMITH 

--7499 ALLEN 

--7521 WARD 

--7654 MARTIN 

--7844 TURNER 

--7900 JAMES 

--7934 MILLER 

--7876 ADAMS 

--7566 JONES 

----7788 SCOTT 

------7876 ADAMS 

----7902 FORD 

------7369 SMITH 

--7698 BLAKE 

----7499 ALLEN 

----7521 WARD 

----7654 MARTIN 

...... 

...... 

39 rows selected. -- Still I got more records 



2 ename From (Select * from test_emp where company_cd=101) test_emp 

3 start with mgr is null -- Change condition 



EMPNO ENAME 

--------------- ---------- 

--7839 KING 

----7566 JONES 

------7788 SCOTT 

--------7876 ADAMS 

------7902 FORD 

--------7369 SMITH 

----7698 BLAKE 

------7499 ALLEN 

------7521 WARD 

------7654 MARTIN 

------7844 TURNER 

------7900 JAMES 

----7782 CLARK 

------7934 MILLER 


I got what i want, but is there any other way to do this.. 

Even though my doublt is ... in From clause I have used inline view which contains only 14 records 

but result shows 39 records... Why they are repeating.. If parent key is combination of company_cd, 

mgr then 

how to write connect by clause 

Thans in Adv. 

Sanjeev Vibuthi 

Followup March 1, 2006 - 8am Central time zone: 

well, if you don't have "start with", you start with EACH RECORDS (meaning - there are 14 emps, 

you'll have 14 trees) 

so, what do you want to start with exactly 

They are NOT repeating, you asked for them all - you have 14 "trees"

March 1, 2006 - 9am Central time zone Bookmark | Bottom | Top 

Reviewer: Tom Fox from Cincinnati, OH 

Tom, so what I'm getting out of the above query (repeating, not repeating, whatever) is that the 

START WITH clause _must_ reference the left hand side of the CONNECT BY clause. Is this correct? 

--Tom 


I don't even know what the "left hand side" of a connect by would be, so no. 

start with identifies the set of ROOT NODES. 

connect by builds a hierarchy of child nodes under each root. 



Yeah, I haven't finished my coffee yet. I meant the right hand side in my post above. 

--Tom 


I don't know what the right hand side is either. 

March 1, 2006 - 5pm Central time zone Bookmark | Bottom | Top 


Oh, you know, the right hand side.... Work with me here. :) 

In this query: 

scott@ORA8I.WORLD> select lpad(' ',level*2,' ')||ename ename, empno, mgr 

2 from emp 

3 START WITH MGR IS NULL 


5 / 

What I was trying to ask: Does the START WITH clause (meaning MGR on line 3) have to reference the 

right side of the CONNECT BY clause (line 4, in this case), in order for the tree to be formed 

correctly? 

In other terms, does START WITH always have to reference one of the fields in the CONNECT BY 

clause? 


No, it does not have to. 

your start with simply identifies the rows that start hierarchies. 

You might want a tree for every MGR that has the JOB = 'X' 

so the start with would be referencing the EMPNO column and the JOB column (to find empnos that are 

mgrs and have a job = 'X' ) 

Can we add salary to the query? March 2, 2006 - 1pm Central time zone Bookmark | Bottom | Top 

Reviewer: Mike from Dallas, TX 

Hi Tom,

Could you show how to add Salary to this query using the connect by? for example I would like to 

add a column to the query that shows the total salary for each manager and his/her employees 

including all employees down to the last leaf. 

In you example above your output shows 


--------------- ---------- ---------- 

KING 7839 

JONES 7566 7839 

SCOTT 7788 7566 

ADAMS 7876 7788 

FORD 7902 7566 

SMITH 7369 7902 

In other words the overall salary column for King would be the sum of 

KING+JONES+SCOTT+ADAMS+FORD+SMITH because they all fall under KING or an employee of KING, the 

overall salary column for SCOTT would be SCOTT's salary + ADAMS and FORD's would be FORD's salary + 

SMITH's 

Of course SMITH's and ADAM's salaries would only include theirs because they don't manage anyone. 

Any help would be greatly appreciated 


ops$tkyte@ORA9IR2> select rpad('*',2*level,'*')||ename nm, empno, mgr, 

2 (select sum(sal) from emp e2 start with e2.empno = emp.empno connect by prior empno = 

mgr) sum_sal 

3 from emp 


5 connect by prior empno = mgr; 

NM EMPNO MGR SUM_SAL 

-------------------- ---------- ---------- ---------- 

**KING 7839 29025 

****JONES 7566 7839 10875 

******SCOTT 7788 7566 4100 

********ADAMS 7876 7788 1100 

******FORD 7902 7566 3800 

********SMITH 7369 7902 800 

****BLAKE 7698 7839 9400 

******ALLEN 7499 7698 1600 

******WARD 7521 7698 1250 

******MARTIN 7654 7698 1250 

******TURNER 7844 7698 1500 

******JAMES 7900 7698 950 

****CLARK 7782 7839 3750 

******MILLER 7934 7782 1300 


Using multiple table join and connect by March 10, 2006 - 3pm Central time zone 



Tom,.. 

Thanks for the quick reply, I apologize for not getting back to you quickly. 

Maybe I didn't explain well enough. a scalar subquery work very well but I am concerned about the 

table growing and causing the query to become very taxing to the system. 

What we have is a table of entities and a table of entity tickets, they are 

CREATE TABLE ENTITY ( 

ENTITY_UUID VARCHAR2(32), 

NAME VARCHAR2(256), 

PARENT_UUID VARCHAR2(32) 

) 

/


CREATE TABLE ENTITY_TCKT ( 

ENTITY_UUID VARCHAR2(32), 

CURRENT_LIFECYCLE_STATE NUMBER 

) 

/ 

To populate these tables and show some stats,.. 

insert into entity values ('13E7CAA5FDEB42518A798A77A19F70B0','Level1 

Entity',NULL); 

insert into entity values ('66A6A6EFFA9D46BE82EC8F5CFFAC91B9','Level4 

Entity','536FCF7E4A5D457B8C3AECBED878FDBF'); 

insert into entity values ('DCF6B6366D6449DB95A5AEA6B14F31F7','Level5 

Entity','66A6A6EFFA9D46BE82EC8F5CFFAC91B9'); 

insert into entity values ('E2FD444948714528805EBFFA102511F5','Level5 

Entity','CB4E1B74035947B9A5B9B0FE264DF4E7'); 

insert into entity values ('2E1E0646AC9F4BB9A6E4A747B20B2595','Level2 

Entity','13E7CAA5FDEB42518A798A77A19F70B0'); 

insert into entity values ('54133391FDD54221B11382A20DFC38AA','Level2 

Entity','13E7CAA5FDEB42518A798A77A19F70B0'); 

insert into entity values ('95A5F85D68184DB7A49F9DF7A236F9AF','Level5 

Entity','99762DC75A5D42DCBEA6950D7011F130'); 

insert into entity values ('EAD30C5578BD491991B0D7049CD4F277','Level4 


insert into entity values ('536FCF7E4A5D457B8C3AECBED878FDBF','Level3 

Entity','54133391FDD54221B11382A20DFC38AA'); 

insert into entity values ('883FD970DF264B7A9DC0DFBAF225012A','Level4 


insert into entity values ('C23E104B5AA044F795A6896B1C0B08E4','Level3 

Entity','54133391FDD54221B11382A20DFC38AA'); 

insert into entity values ('64BA9F3295194A3B955FD446DBB2E7EC','Level4 

Entity','C23E104B5AA044F795A6896B1C0B08E4'); 

insert into entity values ('0FE28D72C40C4D6FBB439A49B0BE6D3F','Level5 

Entity','64BA9F3295194A3B955FD446DBB2E7EC'); 

----------list goes on------------------ 

Then to generate some random ticket data 

DECLARE 

l_tckt_ktr NUMBER; 

l_entity_ktr NUMBER; 

BEGIN 

FOR i IN (select entity_uuid from entity) 

LOOP 

l_entity_ktr := round(dbms_random.value(1,6),0); 

DBMS_OUTPUT.PUT_LINE('Processing entity -> '||i.entity_uuid||' 

l_entity_ktr = '||l_entity_ktr); 

IF l_entity_ktr = 1 THEN 

l_tckt_ktr := round(dbms_random.value(1,10),0); 

FOR n IN 1 .. l_tckt_ktr 

LOOP 

insert into entity_tckt values (i.entity_uuid,0); 

END LOOP; 

ELSIF l_entity_ktr = 3 THEN 



LOOP 


END LOOP; 

ELSIF l_entity_ktr = 5 THEN 



LOOP 


END LOOP; 

END IF; 

END LOOP; 

END; 

/ 

PL/SQL procedure successfully completed. 

Now that the tables have data, try two ways to get the query 

First the Scalar Subquery

set lines 256 

column lpad('',2*level)||e.name format a50 

column lpad('',2*level)||e.entity_uuid format a50 

select lpad(' ', 2 * level)||e.name, lpad(' ', 2 * level)||e.entity_uuid, level, 

(select count(*) from entity_tckt where current_lifecycle_state = 0 and 

entity_uuid in ( 

select a.entity_uuid from entity a start with a.entity_uuid = 

e.entity_uuid connect by prior a.entity_uuid = a.parent_uuid)) as nbr 

from entity e 

start with e.entity_uuid = '13E7CAA5FDEB42518A798A77A19F70B0' connect by prior 

e.entity_uuid = e.parent_uuid 

/ 

LPAD('',2*LEVEL)||E.NAME 

LPAD('',2*LEVEL)||E.ENTITY_UUID LEVEL NBR 

-------------------------------------------------- 

-------------------------------------------------- ---------- ---------- 

Level1 Entity 

13E7CAA5FDEB42518A798A77A19F70B0 1 15533 

Level2 Entity 

2E1E0646AC9F4BB9A6E4A747B20B2595 2 7 

Level2 Entity 

54133391FDD54221B11382A20DFC38AA 2 15526 

Level3 Entity 

536FCF7E4A5D457B8C3AECBED878FDBF 3 890 

Level4 Entity 

66A6A6EFFA9D46BE82EC8F5CFFAC91B9 4 51 

Level5 Entity 

DCF6B6366D6449DB95A5AEA6B14F31F7 5 0 

Level4 Entity 

EAD30C5578BD491991B0D7049CD4F277 4 76 

Level4 Entity 

883FD970DF264B7A9DC0DFBAF225012A 4 0 

Level4 Entity 

0F01178489CB421CA5A4C55AEC98300E 4 4 

Level5 Entity 

5B3B033B4A30443BB1B6F57C1BB05399 5 4 

Level4 Entity 

CB4E1B74035947B9A5B9B0FE264DF4E7 4 69 

LPAD('',2*LEVEL)||E.NAME 

LPAD('',2*LEVEL)||E.ENTITY_UUID LEVEL NBR 

-------------------------------------------------- 

-------------------------------------------------- ---------- ---------- 

Level5 Entity 

E2FD444948714528805EBFFA102511F5 5 69 

Level4 Entity 

99762DC75A5D42DCBEA6950D7011F130 4 607 

Level5 Entity 

95A5F85D68184DB7A49F9DF7A236F9AF 5 517 

-------------------and on and on---------------- 


Execution Plan 

---------------------------------------------------------- 

0 SELECT STATEMENT Optimizer=CHOOSE 

1 0 SORT (AGGREGATE) 

2 1 FILTER 

3 2 TABLE ACCESS (FULL) OF 'ENTITY_TCKT' 

4 2 FILTER 

5 4 CONNECT BY (WITH FILTERING) 

6 5 NESTED LOOPS 

7 6 TABLE ACCESS (FULL) OF 'ENTITY' 

8 6 TABLE ACCESS (BY USER ROWID) OF 'ENTITY' 


10 9 BUFFER (SORT) 

11 10 CONNECT BY PUMP 







18 17 BUFFER (SORT) 


20 17 TABLE ACCESS (FULL) OF 'ENTITY'

Statistics 

---------------------------------------------------------- 

4 recursive calls 

0 db block gets 

208383 consistent gets 

0 physical reads 

0 redo size 

9007 bytes sent via SQL*Net to client 

591 bytes received via SQL*Net from client 

10 SQL*Net roundtrips to/from client 

27913 sorts (memory) 

0 sorts (disk) 

122 rows processed 

And now using 'WITH' 

with alm as (select entity_uuid, count(*) as nbr 

from entity_tckt 

group by entity_uuid) 

select e.entity_uuid, sum(nbr) from entity e, alm 

where alm.entity_uuid in (select entity_uuid from entity start with entity_uuid 

= e.entity_uuid connect by prior entity_uuid = parent_uuid) 

group by e.entity_uuid 

/ 

ENTITY_UUID SUM(NBR) 

-------------------------------- ---------- 

026D4A6B547544E08CFEC5DDED3B7777 772 

02AD7930E9B94A20A9F4E245B3F4B8C4 1747 

04B3AD5F804B4A66AC6C91606EA7019B 74 

070547E30C604D4680089959A6DB7684 677 

08ABA73521F94F2DB19FF8C1C53ADF06 2 

0D65B9BB906B442B9A6BDFEEF3D858AC 86 

0F01178489CB421CA5A4C55AEC98300E 4 

0FE28D72C40C4D6FBB439A49B0BE6D3F 716 

10801AFBCBC44F2F9851E16BC1CCA442 703 

13E7CAA5FDEB42518A798A77A19F70B0 15533 

1677A02C0B89479BA27F95D7AA3DAC03 9516 

ENTITY_UUID SUM(NBR) 

-------------------------------- ---------- 

1AE6BCDA1FA74B0B8DEBEFF13F9A444A 96 

1F9BD4CD72FA4879AC338397FB59FD39 74 

21FAD03A5EB34D79908E96DA647FF24C 592 

235F6C1600584CDD8B3FB1FFCD742E7E 839 

23EA0B4618A94C1586BDF08437B97EB0 3067 

25CFE1E64F36468DB291CBCF0867B314 59 

26C635064E83447B91BA8D125FE74C2A 776 

2ABC45D4A5D84684AC29A578C5CCBD3E 1759 

2DF37BBB64254DB29081F03D38B5CE33 876 

2E1E0646AC9F4BB9A6E4A747B20B2595 7 

2FF85FEF82864C0BB75BA4513885ED0E 76 

----------------more data ---------------- 



---------------------------------------------------------- 

0 SELECT STATEMENT Optimizer=CHOOSE 

1 0 SORT (GROUP BY) 

2 1 FILTER 


4 3 VIEW 

5 4 SORT (GROUP BY) 

6 5 TABLE ACCESS (FULL) OF 'ENTITY_TCKT' 


8 2 FILTER 






14 13 BUFFER (SORT)



Statistics 

---------------------------------------------------------- 

0 recursive calls 

0 db block gets 

197127 consistent gets 

0 physical reads 

0 redo size 

3707 bytes sent via SQL*Net to client 

547 bytes received via SQL*Net from client 

6 SQL*Net roundtrips to/from client 

27902 sorts (memory) 

0 sorts (disk) 

71 rows processed 

The logical reads will kill us as the table grows - Is there another way to get the logical reads 

down. I expect the entity table to grow to several hundred thousand records and the tickets table 

to get very large as well. 


why would you connect by the entire table????? You would never have a where clause? 

single heirarchy with multiple employees March 11, 2006 - 4am Central time zone 


Reviewer: Rasin from UAE 

single heirarchy with multiple employees 

Suppose querying for empno's 7934, 7369 (actually this will be 

determined 

by the input of another subquery) 

and want to get a single heirarchy instead of multiple heirarchies. 

column ename format a60 

select e.empno, lpad(' ', level * 2, ' ') || e.ename ename 

from emp e 

connect by e.empno = prior e.mgr 

start with empno in (7934, 7369) 

EMPNO ENAME 

---------- -------------------------------------------------- 

7369 SMITH 

7902 FORD 

7566 JONES 

7839 KING 

7934 MILLER 

7782 CLARK 

7839 KING 


currently I am doing a distinct select to remove duplicate but I am 

loosing 

the heirarchy 

by doing so and also the subquery which will return empno's can return 

100's of employees, 

will this be OK performance wise ? 

select distinct * 

from 

( 

select e.empno, e.ename ename 

from emp e 

connect by e.empno = prior e.mgr 

start with empno in (7934, 7369) 

order siblings by empno 

)

EMPNO ENAME 

---------- -------------------------------------------------- 

7369 SMITH 

7566 JONES 

7782 CLARK 

7839 KING 

7902 FORD 

7934 MILLER 

My motto is to get the employees suppose whose salary 

is greater than 3000 along with their managers up to the root of the heirarchy. 



will it be OK performance wise? You gotta do what you gotta do and if in fact you want to "start 

with" hundreds of roots, expand them all of the way up and distinct them - that is what you gotta 

do. 

single heirarchy March 12, 2006 - 3pm Central time zone Bookmark | Bottom | Top 

Reviewer: Rasin from UAE 

It took time to get some hold on heirarchial queries. 

My data model looks like following 

DEPT 1:M EMP 

EMP 1:1 PROJECTS (there can be employees which are not assigned to projects) 

PROJECTS 1:M SCHEDULES 

create table projects (projno number primary key, pname varchar2(20)); 

create table schedules (scheduleno number primary key, projno number 

constraint proj_fk references projects (projno), schedule_name varchar2(50)); 

alter table emp add projno number constraint emp_proj_fk 

references projects; 

insert into projects values (1, 'Engineering'); 

insert into projects values (2, 'Maintenance'); 

insert into schedules values (1, 1, 'Schedule1'); 




--assign project 1 to MILLER 

update emp set projno = 1 

where empno = 7934; 

--assign project 2 to SMITH 

update emp set projno = 2 

where empno = 7369; 

commit; 

/* 

I want to get employee records along with their managers 

even though the managers are not assigned any projects. 

The filtering to be done on schedules.schedule_name with in operator 

for eg., I want to get employees for schedule1 and schedule3 along 

with their managers and the heirarchy above them 

*/ 

--I tried the following query with no rows returned 

select e.empno, lpad(' ', level * 2, ' ') || e.ename ename 

, dname, s.schedule_name 

from emp e, dept d, projects p, schedules s 

where 

e.deptno = d.deptno 

and e.projno = p.projno(+)

and p.projno = s.projno 

and s.schedule_name in ('Schedule1', 'Schedule2') 

connect by prior e.empno = e.mgr 

start with e.mgr is null 

--after watching this and other threads and reading oracle documentation 

-- I came up with the following query, I applied your logic of getting 

--the salary based on heirarchy. 

select lpad('*', level * 2, '*') || ename ename, dname, pname 

from 

( 

select e.empno, e.ename ename, e.mgr 

, dname, pname, 

(select count(p2.projno) from emp e2 

,projects p2 

where 

e2.projno = p2.projno(+) 

and 

p2.projno in 

(select projno from schedules 

where schedule_name in ('Schedule1')) 

start with e2.empno = e.empno 

connect by prior empno = mgr) proj_cnt 

from emp e, dept d, projects p 

where 


and e.projno = p.projno(+) 

and 

( 

p.projno in 


where schedule_name in ('Schedule1')) 

or 

e.projno is null 

) 

) 

where proj_cnt > 0 

start with mgr is null 

connect by prior empno = mgr 

/ 

ENAME DNAME PNAME 

------------------------------ -------------- -------------------- 

**KING ACCOUNTING 

****CLARK ACCOUNTING 

******MILLER ACCOUNTING Engineering 

select lpad('*', level * 2, '*') || ename ename, dname, pname 

from 

( 

select e.empno, e.ename ename, e.mgr 

, dname, pname, 

(select count(p2.projno) from emp e2 

,projects p2 

where 

e2.projno = p2.projno(+) 

and 

p2.projno in 


where schedule_name in ('Schedule3','Schedule1')) 

start with e2.empno = e.empno 

connect by prior empno = mgr) proj_cnt 

from emp e, dept d, projects p 

where 


and e.projno = p.projno(+) 

and 

( 

p.projno in 


where schedule_name in ('Schedule3','Schedule1')) 

or 

e.projno is null 

) 

) 

where proj_cnt > 0 

start with mgr is null

connect by prior empno = mgr 

/ 

ENAME DNAME PNAME 

------------------------------ -------------- -------------------- 

**KING ACCOUNTING 

****CLARK ACCOUNTING 

******MILLER ACCOUNTING Engineering 

****JONES RESEARCH 

******FORD RESEARCH 

********SMITH RESEARCH Maintenance 

--clean up 

drop table schedules; 

alter table emp drop column projno; 

drop table projects; 

Won't connect by entire table March 13, 2006 - 5pm Central time zone Bookmark | Bottom | Top 


Tom,.. 

I didn't intend to connect by the entire table, only in this exercise because the tickets table 

only has records were interested in, the ticket table does have a current_lifecycle_state column to 

it. 

Also, not every entity has tickets. The first query show this where clause. Sorry I didn't have it 

in the second query. 

The task I am trying to accomplish is not to have the query go completely down the entity table 

using connect by for every record resulting from the outer connect by. For Example,.. 

Level1- 

|_ 

| Level2_ 

| |_Level3 

| |_Level3 

| |_Level3 

|_Level2_ 

| |_Level3 

| |_Level3_ 

| |_Level4 

| |_Level4 

|_Level2_ 

For each entity, there may/may not be opened tickets in the ticket table. One could create a view 

on tickets for each entity easily using the connect by and scalar subquery that returns the number 

of open tickets for that entity alone. 

Would there be a better way to sum up these numbers from the bottom up and therefore get a faster 

return using less resources than having to use the connect by in the scalar subquery. 

I hope I was a bit clearer. 

Mike 

Please Help May 6, 2006 - 5pm Central time zone Bookmark | Bottom | Top 

Reviewer: Isam from Jordan 

Hi Tom, 

How can I query all Employees ( lets say whose sal >=3000 ) and their managers. 

I used the follwing but the result is not what I want 

select EMPNO,SUBSTR(rpad('*',2*level,'*' ) || ENAME,1,30), MGR,SAL FROM EMP 

START WITH MGR IN (SELECT X.EMPNO FROM EMP X WHERE X.SAL>=3000) 

CONNECT BY PRIOR EMPNO = MGR 

The Result is :

EMPNO SUBSTR(RPAD('*',2*LEVEL,'*')|| MGR SAL 

---------- ------------------------------ ---------- ---------- 

7876 **ADAMS 7788 1100 

7566 **JONES 7839 2975 

7788 ****SCOTT 7566 3000 

7876 ******ADAMS 7788 1100 

7902 ****FORD 7566 3000 

7369 ******SMITH 7902 800 

7698 **BLAKE 7839 2850 

7499 ****ALLEN 7698 1600 

7521 ****WARD 7698 1250 

7654 ****MARTIN 7698 1250 

7844 ****TURNER 7698 1500 


---------- ------------------------------ ---------- ---------- 

7900 ****JAMES 7698 950 

7782 **CLARK 7839 2450 

7934 ****MILLER 7782 1500 

7369 **SMITH 7902 800 


Thank you 

Followup May 7, 2006 - 11am Central time zone: 

"query all Employees ( lets say whose sal >=3000 ) and their managers." 

why would you use connect by? 

ops$tkyte@ORA10GR2> select ename, mgr, sal from emp where sal >= 3000; 

ENAME MGR SAL 

---------- ---------- ---------- 

SCOTT 7566 3000 

KING 5000 

FORD 7566 3000 

or if you needed their manager name: 

1 select a.ename, b.ename mgr, a.sal 

2 from emp a, emp b 

3 where a.sal >= 3000 

4* and a.mgr = b.empno(+) 

ops$tkyte@ORA10GR2> / 

ENAME MGR SAL 

---------- ---------- ---------- 

FORD JONES 3000 

SCOTT JONES 3000 

KING 5000 

May 7, 2006 - 5pm Central time zone Bookmark | Bottom | Top 


Thank you for reply, the idea is to get all employees who satisfies the condition (sal>=3000) and 

at the same time I want to know their managers of all levels regardless of manager's sal in a tree 

structure. the result I am looking for is like this: 

EMPNO name MGR SAL 

---------- ------------------------------ ---------- ---------- 

7839 **KING 5000 

7566 ****JONES 7839 2975 

7788 ******SCOTT 7566 3000 

7902 ******FORD 7566 3000

7902 ******FORD 7566 3000 

as you can see , the condition returns King, Scott and Ford . but Jones in the list because he is 

the MGR of Scott and Ford. 

That what I want. How can I do this. 


Followup May 8, 2006 - 7am Central time zone: 

that is a bit of a sticky problem. we can do it upside down rather easily: 

ops$tkyte@ORA10GR2> select EMPNO,SUBSTR(rpad('*',2*level,'*' ) || ENAME,1,30), MGR,SAL 

2 FROM EMP 

3 start with sal >= 3000 

4 CONNECT BY PRIOR mgr = empno 

5 / 


---------- ------------------------------ ---------- ---------- 

7788 **SCOTT 7566 3000 

7566 ****JONES 7839 2975 

7839 ******KING 5000 

7839 **KING 5000 

7902 **FORD 7566 3000 

7566 ****JONES 7839 2975 

7839 ******KING 5000 


but going from the top down is a tad harder: 

ops$tkyte@ORA10GR2> select EMPNO,SUBSTR(rpad('*',2*level,'*' ) || ENAME,1,30), MGR,SAL 

2 FROM EMP 

3 where empno in ( select EMPNO 

4 FROM EMP 


6 CONNECT BY PRIOR mgr = empno ) 

7 start with empno in ( select EMPNO 

8 FROM EMP 

9 where connect_by_isleaf = 1 


11 connect by prior mgr = empno ) 

12 CONNECT BY PRIOR empno = mgr 

13 / 


---------- ------------------------------ ---------- ---------- 

7839 **KING 5000 

7566 ****JONES 7839 2975 

7788 ******SCOTT 7566 3000 

7902 ******FORD 7566 3000 

it would only make sense to do the "start with" subquery IF you have multiple roots and thought 

that the filtering done by the subquery would prune away large parts of the hierarchy - otherwise, 

just build the entire hierarchy and then prune it with the where clause. 

May 9, 2006 - 4pm Central time zone Bookmark | Bottom | Top 


Thank you for help. 

That is what I need. I will use this logic in different queries of my project. 

Thanks again. 

Connect by doing extra full Scan? June 1, 2006 - 12pm Central time zone Bookmark | Bottom | Top 

Reviewer: Greg from Toronto

Hi Tom, 

Got myself confused (again - *sigh*) and was hoping you could shed some light on something. 

I managed to reproduce in this test case: 

gregs-ORA10 > drop table junk; 

Table dropped. 

gregs-ORA10 > create table junk 

2 ( col1 number, 

3 col2 number ) 

4 / 


gregs-ORA10 > insert into junk values ( 1, null ); 

1 row created. 



gregs-ORA10 > insert into junk values ( 3, 2 ); 












gregs-ORA10 > commit; 

Commit complete. 

gregs-ORA10 > set autotrace traceonly explain 

gregs-ORA10 > select col1 

2 from junk 

3 start with col2 = 2 

4 connect by prior col1 = col2 

5 / 


---------------------------------------------------------- 

Plan hash value: 3214713417 

------------------------------------------------------------------------------------- 

| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | 

------------------------------------------------------------------------------------- 

| 0 | SELECT STATEMENT | | 8 | 208 | 2 (0)| 00:00:01 | 

|* 1 | CONNECT BY WITHOUT FILTERING| | | | | | 

|* 2 | TABLE ACCESS FULL | JUNK | 1 | 52 | 2 (0)| 00:00:01 | 

| 3 | TABLE ACCESS FULL | JUNK | 8 | 208 | 2 (0)| 00:00:01 | 

------------------------------------------------------------------------------------- 

Predicate Information (identified by operation id): 

--------------------------------------------------- 

1 - access("COL2"=PRIOR "COL1")

2 - filter("COL2"=2) 

Note 

----- 

- dynamic sampling used for this statement 

gregs-ORA10 > create index junk_ind1 on junk ( col2 ) 

2 / 

Index created. 


2 from junk 



5 / 


---------------------------------------------------------- 


------------------------------------------------------------------------------------------- 


------------------------------------------------------------------------------------------- 


|* 1 | CONNECT BY WITH FILTERING | | | | | | 

| 2 | TABLE ACCESS BY INDEX ROWID | JUNK | 1 | 52 | 1 (0)| 00:00:01 | 

|* 3 | INDEX RANGE SCAN | JUNK_IND1 | 1 | | 1 (0)| 00:00:01 | 

| 4 | NESTED LOOPS | | | | | | 

| 5 | BUFFER SORT | | | | | | 

| 6 | CONNECT BY PUMP | | | | | | 

| 7 | TABLE ACCESS BY INDEX ROWID| JUNK | 1 | 26 | 1 (0)| 00:00:01 | 



------------------------------------------------------------------------------------------- 


--------------------------------------------------- 

1 - access("COL2"=PRIOR "COL1") 

3 - access("COL2"=2) 


Note 

----- 

- dynamic sampling used for this statement 

gregs-ORA10 > analyze table junk compute statistics; 

Table analyzed. 


2 from junk 



5 / 


---------------------------------------------------------- 


------------------------------------------------------------------------------------------- 


------------------------------------------------------------------------------------------- 



| 2 | TABLE ACCESS BY INDEX ROWID | JUNK | 1 | 6 | 1 (0)| 00:00:01 | 


| 4 | NESTED LOOPS | | | | | | 

| 5 | BUFFER SORT | | | | | | 


| 7 | TABLE ACCESS BY INDEX ROWID| JUNK | 2 | 6 | 1 (0)| 00:00:01 | 



------------------------------------------------------------------------------------------- 


---------------------------------------------------


3 - access("COL2"=2) 


This is and Oracle 10g (10.2.0.2.0) database, and my question is, why is Oracle doing that final 

FULL SCAN on JUNK ?? I just don't understand the logic - I don't know where it's coming from. I 

think the "START WITH" results in one Index scan, and the connect by results in the 2nd ... but I 

can't figure out that 3rd scan. 

The original query ran fine in 8i (doing 2 range scans, and no third scan at all!) 

Sorry ... don't mean to harass .. ;) June 5, 2006 - 10am Central time zone Bookmark | Bottom | Top 

Reviewer: Greg from Toronto 

I know you're busy, and I know you don't always see every post ... however, we're really stuck on 

this, I've tried the Oracle Metalink Forums ... nothing yet ... and we've continued to try a few 

other things .. (I'm currently reading through Jonathan's book "Cost Based Oracle Fundamentals" in 

the hopes of sheding some light on it .. but I'm still stumped ... 

If you can find some time in your busy schedule to take a peek at this one, I'd be internally 

grateful ... if you can't ... well .. I understand!! ;) 

Followup June 5, 2006 - 10am Central time zone: 

this is a teeny tiny table, I don't see the point - the table is too small to use the index here. 

True .. however ... June 6, 2006 - 1pm Central time zone Bookmark | Bottom | Top 


This is a re-created test case based on a real problem I have with a table that's 17 million rows 

... 

Same explain is showing up in that query as this one. 

Statistics are all up to date ... 

I can also re-produce the test case with a ~17 million row table .. but I was hoping an 8 row table 

might be easier to work with ?? 

Just to be clear .. it's not that I'm going to complain about the full table scan .. (yet) .. just 

wondering why Oracle has decided to make that third pass ... I can't seem to understand why it 

would want to make a third pass on that data ... all because we added an index ?? 

(Note the explain with no index does 2x full scans ... but after the index is added, we're making 

3x passes ... 2x index scans, + 1x full ... ) 

I'm only trying to understand that third pass .. 

Thanks! 

Followup June 6, 2006 - 2pm Central time zone: 

it is a "dummy no-op" pass, it is an artifact of the explain plan output. 

select col1 

from junk 

start with col2 = 2 

connect by prior col1 = col2 

call count cpu elapsed disk query current rows 

------- ------ -------- ---------- ---------- ---------- ---------- ----------

------- ------ -------- ---------- ---------- ---------- ---------- ---------- 

Parse 1 0.00 0.00 0 0 0 0 

Execute 1 0.00 0.00 0 0 0 0 

Fetch 2 0.00 0.00 0 5 0 2 

------- ------ -------- ---------- ---------- ---------- ---------- ---------total 

4 0.00 0.00 0 5 0 2 

Misses in library cache during parse: 1 

Optimizer mode: ALL_ROWS 

Parsing user id: 140 

Rows Row Source Operation 

------- --------------------------------------------------- 

2 CONNECT BY WITH FILTERING (cr=5 pr=0 pw=0 time=239 us) 

1 TABLE ACCESS BY INDEX ROWID JUNK (cr=2 pr=0 pw=0 time=61 us) 

1 INDEX RANGE SCAN JUNK_IND1 (cr=1 pr=0 pw=0 time=40 us)(object id 

1 NESTED LOOPS (cr=3 pr=0 pw=0 time=113 us) 

2 BUFFER SORT (cr=0 pr=0 pw=0 time=65 us) 

2 CONNECT BY PUMP (cr=0 pr=0 pw=0 time=29 us) 

1 TABLE ACCESS BY INDEX ROWID JUNK (cr=3 pr=0 pw=0 time=37 us) 

1 INDEX RANGE SCAN JUNK_IND1 (cr=2 pr=0 pw=0 time=24 us)(object id 

0 TABLE ACCESS FULL JUNK (cr=0 pr=0 pw=0 time=0 us) 

see the zero rows, cr=0, no activity - do you have a tkprof where the row source operation shows 

"not zero" 

yes ... but ... June 6, 2006 - 8pm Central time zone Bookmark | Bottom | Top 


Yeah .. but I'm not sure we're looking at the same thing! ;) 

In my original post, the set autotrace traceonly explain command spit out an explain formatted a 

bit different than yours (which is why I'm thinking I'm not looking at the right thing .. hehe) ... 

and it shows 8 rows on this one: 


I can try to re-run with: 

set autotrace traceonly statistics ?? 

Followup June 6, 2006 - 9pm Central time zone: 

that is an autotrace, it is not "what happened". 

tkprof shows "what happened" 

Here is more information: 

... 

The connect by row source uses a sort to store the rows it will be working on. If the filtering 

option is used, connect by needs the sort to detect duplicates as the rows are inserted into the 

sort. If the sort spills to disk, it can no longer detect duplicates at the time the rows are 

inserted (the duplicates will be detected later, when the sort runs are merged). Therefore if the 

sort spills to disk, the connect by will switch to "no filtering" mode. The extra line in the plan 

is the row source that will be used if the switch to "no filtering" happens. 

............. 

Tree Walking from the result of a join June 14, 2006 - 2pm Central time zone 


Reviewer: Mike Jones from England 

Help! I'm trying to write a Query where given a result set, one of the columns is an index into a 

hierarchy. I want to report the result set and then each of the corresponding parents back up the 

hierachy, sort of like reporting some information about a person and then each of their dad, and 

then that person and their dads dad etc. all the way back up.

I can't seem to get it working, I can't get the filtering to the correct tree-walk right. 

Hopefully the below illistrates this. I should only get 6 rows back, the 2 base rows from the ilv 

and these reported 3 times each for the 3 levels of the hierachy. In practice the ILV rows would 

report different IDX values and so the relative tree walks would differ. Can you help point out 

where I'm going mad? 


Mike. 

create table hierarchy_table as 

select object_name, rownum idx, decode(rownum -1,0,null,rownum-1) parent_idx 

from user_objects 

where rownum < 4 

/ 

create table base_x as 

select 3 idx, object_name 

from user_objects 

where rownum < 3 

/ 

create table base_y as 

select * 

from base_x 

/ 

select ilv.object_name, ilv.idx ilv_idx, ht.object_name ht_name, level 

from 

( select x.object_name, x.idx 

from base_x x, base_y y 

where x.idx = y.idx 

) ilv, 

hierarchy_table ht 

connect by prior ht.parent_idx = ht.idx 

start with ilv.idx = ht.idx 

order by 1 

/ 


I'm not sure why this isn't just a normal connect by - your text description is that of a connect 

by: 

... 

I want to report the result set and then each of 

the corresponding parents back up the hierachy, 

..... 

(the result set = START WITH rows, corresponding parents = CONNNECT BY rows) 

so, maybe if you show us what you get and what you think you should get I can understand better. 

Connect by Join Bug? August 9, 2006 - 10am Central time zone Bookmark | Bottom | Top 

Reviewer: Matt Turner from AA 

Tom I have produced a sample TEST case, can you tell me if you think this is working as the manual 

or is a bug with the join and connect by: 

We have a user_state table which stores a script number for a connected user(simplified): 

CREATE TABLE user_state (username VARCHAR2(30), scriptid NUMBER ); 

Both Fred and Bill are running script 1. 

INSERT INTO user_state VALUES('FRED',1); 

INSERT INTO user_state VALUES('BILL',1); 

We have a second table - script structure, which has the script id, and a list of questions (TREE).

INSERT INTO script_structure VALUES (1,1,NULL); 

INSERT INTO script_structure VALUES (1,2,1); 



So, for Script 1: 

SELECT * 

FROM script_structure 

WHERE scriptid = 1 

START WITH parent_que_id IS NULL 

CONNECT BY PRIOR que_id = parent_que_id 

AND PRIOR scriptid = scriptid; 

1 1 

1 2 1 

1 3 2 

1 4 2 

Looks ok so far. Now I want to join to the user_state table. 

SELECT * 

FROM script_structure ss, 

user_state us 

WHERE username = 'FRED' /* Derived from a DB context */ 

AND us.scriptid = ss.scriptid 


CONNECT BY PRIOR que_id = parent_que_id; 

This is where it goes pear shaped with the result set: 

1 1 FRED 1 

1 2 1 FRED 1 

1 3 2 FRED 1 

1 4 2 FRED 1 

1 3 2 FRED 1 

1 4 2 FRED 1 

1 2 1 FRED 1 

1 3 2 FRED 1 

1 4 2 FRED 1 

1 3 2 FRED 1 

1 4 2 FRED 1 

I do have a work around, that involves using an inline view to restrict the dataset prior to the 

tree walk. 

SELECT que_id, parent_que_id 

FROM 

( 

SELECT * 

FROM script_structure ss, 

user_state us 

WHERE username = 'FRED' 

AND us.scriptid = ss.scriptid ) 


CONNECT BY PRIOR que_id = parent_que_id; 

1 

2 1 

3 2 

4 2 

So it seems that the username filter predicate isn't being applied at the time of the join. 

Is this a bona-fide bug?

Followup August 9, 2006 - 11am Central time zone: 

with the connect by and joins: 

join is done 

start with applied 

connect by done 

where clause applied to results 

username = 'FRED' is a predicate, applied after the hierarchy is built. 

time zone 

Tom Can you please tell me the workaround for Connect By Loop Error August 14, 2006 - 8am Central 

Reviewer: Nitin Joshi from INDIA 

Hi Tom, 


Can you please tell is there any workaround to avoid connect By loop like 'NOCYCLE' the one which 

we have in 10g 

Regards 

Nitin S. Jsohi 


only if you can encode it in the query (eg: describe to me what condition would always exist for 

you to have a connect by) 

then we can possibly filter it in the connect by clause (eg: if there is some "rule") 

for example, if you hit the same ID as you started with, you should "stop" (eg: all of your data is 

a "circle") 

ops$tkyte%ORA9IR2> create table t ( id number, pid number ); 


ops$tkyte%ORA9IR2> 

ops$tkyte%ORA9IR2> insert into t values ( 1, 5 ); 











ops$tkyte%ORA9IR2> select * 

2 from t 

3 start with id = 1 

4 connect by prior id = pid; 

ERROR: 

ORA-01436: CONNECT BY loop in user data



ops$tkyte%ORA9IR2> select * 

2 from t 


4 connect by prior id = pid and id 1; 

ID PID 

---------- ---------- 

1 5 

2 1 

3 2 

4 3 

5 4 

but short of having some "logic" or "rule" that can be used to describe when you would have a 

loop.... 

Connect By August 26, 2006 - 5pm Central time zone Bookmark | Bottom | Top 

Reviewer: Saeed from jordan 

Hi Tom , 

I have the following table and its data; 

CREATE TABLE MyTable 

(MyPK NUMBER(3)NOT NULL, 

COL VARCHAR2(20)NOT NULL, 

STATUS NUMBER(1) DEFAULT 1 NOT NULL, 

PARENT_ID NUMBER(3) 

); 

*** Status has one of two values (1=Enabled,0=Disabled) 

ALTER TABLE Mytable ADD ( 

CONSTRAINT StatusCons CHECK (STATUS IN (0, 1))); 


CONSTRAINT MyTable_PK PRIMARY KEY (MyPK)); 


CONSTRAINT MyTable_MyTable_FK FOREIGN KEY (PARENT_ID) 

REFERENCES Mytable (MyPK)); 

INSERT INTO Mytable ( MyPK, col, STATUS, PARENT_ID ) VALUES ( 

0, '0_', 1, NULL); 


1, '1_0', 1, 0); 


2, '2_0', 1, 0); 


3, '3_1', 1, 1); 


4, '4_1', 1, 1); 


5, '5_2', 1, 2); 


6, '6_2', 1, 2); 


7, '7_0', 1, 0); 


8, '8_0', 0, 0); 


9, '9_0', 0, 0); 


10, '10_7', 1, 7); 


11, '11_7', 1, 7); 


12, '12_7', 1, 7); 


13, '13_12', 0, 12); 


14, '14_12', 0, 12); 

INSERT INTO Mytable ( MyPK, col, STATUS, PARENT_ID ) VALUES (

15, '15_11', 0, 11); 


16, '16_15', 1, 15); 

COMMIT; 

MY QUESTION IS : 

How can I get rows that satisfy the following conditions : 

All children with STATUS=1 AND their parents have STATUS=1. 

For Parents with status=0 , don't get their children (regardless of Children's STATUS). 

for Children with Status=0 , don't get them. 

Thanks for help 

Saeed 


do you just want "root" parents - or all parents (eg: what is the start with here) 

August 29, 2006 - 6am Central time zone Bookmark | Bottom | Top 

Reviewer: Saeed from Jordan 

Hi Tom , 

What I want is (all Parents having STATUS=1 and their children having STATUS=1 ) in a tree 

structure. 

i.e All Rows selected must have STATUS=1. 

(DON'T INCLUDE ANY CHILD HIS PARENT'S STATUS=0.) 

I know my English Language is not good enough , but I hope you got the idea. 

Thanks again 

Saeed 


just add "and status = 1" to the connect by clause then, if status 1, it will stop traversing 

the tree. 

August 29, 2006 - 4pm Central time zone Bookmark | Bottom | Top 

Reviewer: Saeed from Jordan 

Thanks TOM , 

It's worked well 

Multiple records within a hierarchy. October 10, 2006 - 11am Central time zone 



Is it possible to get multiple records of a particular employee within a hierarchy in specific 

cases? 

I have the following table structure. 

create table EMPINFO 

( c_STATUS VARCHAR2(2), 

c_GROUP VARCHAR2(10), 

c_SYSID VARCHAR2(10), 

c_SYSTEM VARCHAR2(10), 

c_NAME VARCHAR2(40),

c_NAME VARCHAR2(40), 

c_ID VARCHAR2(10), 

c_MGRID VARCHAR2(10), 

c_MGRNAME VARCHAR2(40) 

); 

Typical Data is 

insert into EMPINFO VALUES ('A','HR','','','JASON BICKER','0000001','',''); 

insert into EMPINFO VALUES ('A','HR','','','NANCY WALTER','0400001','0000001','JASON BICKER'); 

insert into EMPINFO VALUES ('','MF','NAN3005','WIN','NANCY WALTER','0400001','',''); 

insert into EMPINFO VALUES ('','MF','NAN3005','UNIX','NANCY WALTER','0400001','',''); 

insert into EMPINFO VALUES ('A','HR','','','ADAM BRYAN','0400002','0000001','JASON BICKER'); 

insert into EMPINFO VALUES ('','MF','BRY0034','WIN','ADAM BRYAN','0400002','',''); 

insert into EMPINFO VALUES ('A','HR','','','STING CORY','0400003','0400002','ADAM BRYAN'); 

insert into EMPINFO VALUES ('','MF','STI4040','UNIX','STING CORY','0400003','',''); 

For example I want to display all records of NANCY WALTER. The c_MGRID is blank for 2nd and 3rd 

record of NANCY WALTER. Similar structure will follow for other records as well. 

Regards, 

Rao 


if nancy walker doesn't have a manager, how is "she" in the hierarchy???? 



This table is kind of a master table that includes employee and its manager and the 

systems(Unix/Windows) and userids the employee has on it. 

Nancy Walker has a manager whose c_MGRID is populated for the 1st row. But the 2nd and 3rd row of 

Nancy Walker shows 

the system and the userid's she has access to. c_MGRID of this 2nd and 3rd row is empty. Is it 

possible to include these rows in the hierarchy as well similar to "start with .... connect by 

prior" 


need more details, can there be two (or zero, or more than two) records for nancy in there with 

c_mgrid filled in? would they have the same value? 

October 11, 2006 - 4pm Central time zone Bookmark | Bottom | Top 


For a user 

a) only one record wherver c_mgrid is filled. This will always be there when he/she joins the 

company. 

b) 0 or more records wherever c_mgrid is not filled. Rows will be there only if user has access to 

any Windows/Unix/Mainframe systems and has an userid on it. 


use this as your "source" to connect by on: 

ops$tkyte%ORA10GR2> select c_name, max( c_mgrid) over (partition by c_name) mgrid 

2 from empinfo; 

C_NAME MGRID 

---------------------------------------- ---------- 

ADAM BRYAN 0000001 

ADAM BRYAN 0000001


JASON BICKER 

NANCY WALTER 0000001 



STING CORY 0400002 





I am sorry I did not get you. The order does not seem right. 

select c_name, max( c_mgrid) over (partition by c_name) mgrid from empinfo 

start with c_id='0000001' 

connect by prior c_id=c_mgrid 

The O/p is 

C_NAME MGRID 

------------------------------ 


JASON BICKER 



The required O/p is 

C_NAME MGRID 

------------------------------ 

JASON BICKER 

NANCY WALTER '0000001' 

NANCY WALTER '0000001' ->> Access to Windows System 

NANCY WALTER '0000001' ->> Access to unix System 

ADAM BRYAN '0000001' 

ADAM BRYAN '0000001' ->> Access to Windows 

STING CORY '0400002' 

STING CORY '0400002' ->> Access to unix 


no, use my query as your "table 

select .. 

from (MY_QUERY) 

start with 




The ' ' in the C_MGRID section of Required O/p is unintentional. It should be 

The required O/p is 

C_NAME MGRID 

------------------------------ 

JASON BICKER 


NANCY WALTER 0000001 ->> Access to Windows System 

NANCY WALTER 0000001 ->> Access to unix System 


ADAM BRYAN 0000001 ->> Access to Windows 


STING CORY 0400002 ->> Access to unix 

October 12, 2006 - 3pm Central time zone Bookmark | Bottom | Top 

Reviewer: A reader

Thanks a lot for the information. 

CONNECT BY October 27, 2006 - 1pm Central time zone Bookmark | Bottom | Top 

Reviewer: lore from Mexico 

The ansewer is too clear and has an example that clarify the use of the connect by 

Connect by February 15, 2007 - 5pm Central time zone Bookmark | Bottom | Top 

Reviewer: guest from India 

Hi Tom, 

I need output like given below 

Key2 Key1 ENAME 

---------- ------------------------------ 

1 1 **FRENCH 

1 2 ****BLAKE 

1 3 ******ADAMS 

2 1 **FRENCH 

2 2 ****DAVIS 

2 3 ******ADAMS 

3 1 **FRENCH 

3 2 ****OLIVER 

4 1 **BALMAR 

4 2 ****GOWD 

Could you please give me a query to get the above result. with key1 and key2 and tree should have full start to stop. 

eg: 

A->B->D 

A->C->E 

F->G->H 

I->J 

I->K 

Database is 10g. 


Followup February 16, 2007 - 1pm Central time zone: 

hmmm 

eh? 

geez..... 

connect by February 16, 2007 - 2pm Central time zone Bookmark | Bottom | Top 

Reviewer: Guest from UK 

Hi Tom, 

I have written the below code to achieve a hierarchy tree. It works fine but takes lot of time and memory. sometimes sorting 

fails. 

Could you please tell me how can i reduce the run time for the below proc. None of the table has indexes. it has to process 

millions of records to give the tree with parent child relationship. 

Example Input: 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (1, 'C','GBP','DW','A'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (2, 'C','GBP','DW','C'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (3, 'C','GBP','DW','G'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (4, 'C','GBP','DW','F'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (5, 'C','GBP','DW','H'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (6, 'C','GBP','DW','I'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (7, 'C','GBP','DW','J'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (8, 'C','GBP','DW','L'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (9, 'C','GBP','DW','K');

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (9, 'C','GBP','DW','K'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (10, 'C','GBP','DW','P'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (11, 'C','GBP','DW','D'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (12, 'C','GBP','DW','M'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (13, 'C','GBP','DW','O'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (14, 'C','GBP','DW','Q'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (15, 'C','GBP','DW','N'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (16, 'C','GBP','DW','E'); 

insert into po(po_NUM,mat_code,op_plant, sc, batch_num ) values (17, 'C','GBP','DW','B'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (1, 'GBP','DW', 'C','C', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (1, 'GBP','DW', 'C','D', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (1, 'GBP','DW', 'C','E', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (2, 'GBP','DW', 'C','F', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (2, 'GBP','DW', 'C','G', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (4, 'GBP','DW', 'C','H', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (5, 'GBP','DW', 'C','I', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (5, 'GBP','DW', 'C','J', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (7, 'GBP','DW', 'C','L', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (7, 'GBP','DW', 'C','K', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (7, 'GBP','DW', 'C','M', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (9, 'GBP','DW', 'C','P', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (11, 'GBP','DW', 'C','J', 'GBP'); 



insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (16, 'GBP','DW', 'C','O', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (16, 'GBP','DW', 'C','N', 'GBP'); 

insert into pod(po_NUM,op_plant, sc, i_mat_code, ib, ip_plant) values (17, 'GBP','DW', 'C','E', 'GBP'); 

expected output: 

M--C--GBP 1 1 2 16/02/2007 19:25:18 16/02/2007 19:25:18 

Q--C--GBP 1 1 1 16/02/2007 19:25:18 16/02/2007 19:25:18 

N--C--GBP 2 1 3 16/02/2007 19:25:18 16/02/2007 19:25:18 

E--C--GBP 2 1 2 16/02/2007 19:25:18 16/02/2007 19:25:18 

B--C--GBP 2 1 1 16/02/2007 19:25:18 16/02/2007 19:25:18 

M--C--GBP 2 2 4 16/02/2007 19:25:18 16/02/2007 19:25:18 

O--C--GBP 2 2 3 16/02/2007 19:25:18 16/02/2007 19:25:18 

E--C--GBP 2 2 2 16/02/2007 19:25:18 16/02/2007 19:25:18 

B--C--GBP 2 2 1 16/02/2007 19:25:18 16/02/2007 19:25:18 

I--C--GBP 3 1 5 16/02/2007 19:25:19 16/02/2007 19:25:19 

H--C--GBP 3 1 4 16/02/2007 19:25:19 16/02/2007 19:25:19 

F--C--GBP 3 1 3 16/02/2007 19:25:19 16/02/2007 19:25:19 

C--C--GBP 3 1 2 16/02/2007 19:25:19 16/02/2007 19:25:19 

A--C--GBP 3 1 1 16/02/2007 19:25:19 16/02/2007 19:25:19 

P--C--GBP 3 2 7 16/02/2007 19:25:19 16/02/2007 19:25:19 

K--C--GBP 3 2 6 16/02/2007 19:25:19 16/02/2007 19:25:19 

J--C--GBP 3 2 5 16/02/2007 19:25:19 16/02/2007 19:25:19 

H--C--GBP 3 2 4 16/02/2007 19:25:19 16/02/2007 19:25:19 

F--C--GBP 3 2 3 16/02/2007 19:25:19 16/02/2007 19:25:19 

C--C--GBP 3 2 2 16/02/2007 19:25:19 16/02/2007 19:25:19 

A--C--GBP 3 2 1 16/02/2007 19:25:19 16/02/2007 19:25:19 

L--C--GBP 3 3 6 16/02/2007 19:25:19 16/02/2007 19:25:19 

J--C--GBP 3 3 5 16/02/2007 19:25:19 16/02/2007 19:25:19 

H--C--GBP 3 3 4 16/02/2007 19:25:19 16/02/2007 19:25:19 

F--C--GBP 3 3 3 16/02/2007 19:25:19 16/02/2007 19:25:19 

C--C--GBP 3 3 2 16/02/2007 19:25:19 16/02/2007 19:25:19 

A--C--GBP 3 3 1 16/02/2007 19:25:19 16/02/2007 19:25:19 

56 rows. 

total 56 rows expected for the above input. 

The procedure is given below. Please help me to minimize the below query to execute fast and give the result. 

CREATE OR REPLACE PROCEDURE hierarchy_Proc IS

CREATE OR REPLACE PROCEDURE hierarchy_Proc IS 

tempvar varchar2(255); 

batch_cnt number :=1; 

sort_cnt number; 

cnt number; 

old_rec varchar2(255):='TRY'; 

BranchNoV number :=1; 

Key1V number :=1; 

Key2V number :=1; 

InputV varchar2(255); 

ibiv varchar2(255); 

obiv varchar2(255); 

bgsseq number; 

minsort number; 

bgs_chk number; 

outbatch number; 

row_chk number; 

cursor otb_id is select obi from oo; 

cursor ob_id is select obi from oo where obi in (select obi from oo GROUP BY obi having count(*) > 1); 

BEGIN 

-- P 1 

execute immediate 'truncate table bio'; 

execute immediate 'truncate table oo'; 

execute immediate 'truncate table oi'; 

-- Load the bio table with data from the po table 

insert into bio (select (po.batch_num||'--'||po.mat_code||'--'||po.op_plant) obi, (pod.input_batch||'--'||pod.i_mat_code||'-- 

'||pod.ip_plant) ibi, po.load_date,po.update_date 

from po, pod 

where pod.order_num=po.order_num 

and pod.op_plant=po.op_plant ); 

commit; 

-- Remove Circular data from bio table 

delete from bio where obi in(select a.obi from 

bio a inner join bio b 

on a.ibi=b.obi and a.obi=b.ibi ); 

commit; 

--Remove duplicate data from bio table 

delete from bio where rowid not in 

(SELECT MIN(rowid) 

FROM bio 

GROUP BY obi, ibi); 

commit; 

-- For all records where the ibi does not exist as an obi 

For rec in (select distinct(ibi) from (select a.ibi from bio a minus select b.obi from bio b)) 

loop 

insert into bio values(rec.ibi,null,to_date(sysdate,'dd/mm/yyyy'),to_date(sysdate,'dd/mm/yyyy')); 

end loop; 

commit; 

insert into oo(sort_num,obi,ibi,load_date,update_date) (select rownum,obi,ibi,load_date,update_date from bio); 

commit; 

-- load oi table 

insert into oi(sort_num,obi,oi,load_date,update_date)(select sort_num,obi,sort_num,load_date,update_date from oo ); 

commit; 

-- load oi column in the oi table for obi 

For ob_rec in ob_id 

loop 

select min(sort_num) into minsort from oo where obi =ob_rec.obi; 

update oi set oi=minsort where obi=ob_rec.obi; 

end loop; 

commit; 

-- load o_count column in the oi table for obi 

cnt:=1; 

sort_cnt:=1;

sort_cnt:=1; 

For otb_rec in otb_id loop 

if(old_rec = otb_rec.obi) then 

cnt:=cnt+1; 

update oi set o_count=cnt where obi =otb_rec.obi and sort_num=sort_cnt; 

sort_cnt:=sort_cnt+1; 

else 

cnt:=1; 

update oi set o_count=cnt where obi=otb_rec.obi and sort_num=sort_cnt; 

sort_cnt:=sort_cnt+1; 

end if; 

old_rec:=otb_rec.obi; 

end loop; 

commit; 

--Phase 2 

execute immediate 'truncate table bgs'; 

execute immediate 'truncate table bgr'; 

 

For eocbid in ( select distinct(obi) from bio where obi not in (select nvl(ibi,0) from bio)) loop 

obiv := eocbid.obi; 

 

select count(*) into bgs_chk from bgs where obi=obiv; 

if (bgs_chk=0) then 

goto label3; 

else 

goto label4; 

end if; 

 

bgsseq :=1; 

BranchNoV := bgsseq; 

insert into bgs(obi,seq_num,load_date,update_date) values (obiv,bgsseq,sysdate,sysdate); 

commit; 

goto label5; 

 

select count(obi) into outbatch from bgs where obi=obiv; 

if(outbatch=0) then 

bgsseq :=1; 

BranchNoV := bgsseq; 

insert into bgs(obi,seq_num,load_date,update_date) values (obiv,bgsseq,sysdate,sysdate); 

commit; 

else 

select bgs.seq_num+1 into BranchNoV from bgs bgs where bgs.obi=obiv; 

end If; 

 

select count(*) into row_chk from oi oi where oi.obi=obiv and oi.o_count=BranchNoV; 

If(row_chk0) then 

select ibi into InputV from oo oo where oo.sort_num=(select distinct(oi.oi + BranchNoV - 1) from oi oi where oi.obi= obiv); 

If InputRetV is null then 

insert into bgr(batch_num,Key_1,Key_2,gen_sequence,load_date,update_date) (SELECT obi, Key1V,Key2V, (select 

count(*) from bgs)+1-LEVEL,sysdate,sysdate FROM bgs start with ibi is null CONNECT BY PRIOR obi= ibi ); 

 

ibiv:=obiv; 

commit; 

delete from bgs bgs where bgs.obi= obiv; 

select obi into obiv from bgs bgs where bgs.ibi =ibiv ; 

Key2V := Key2V + 1; 

commit; 

goto label2; 

else 

update bgs bgs set bgs.ibi=InputV, bgs.seq_num = BranchNoV,update_date=sysdate where bgs.obi = obiv;

update bgs bgs set bgs.ibi=InputV, bgs.seq_num = BranchNoV,update_date=sysdate where bgs.obi = obiv; 

obiv := InputRetV; 

commit; 

goto label2; 

end If; 

else 

ibiv := obiv; 

delete from bgs bgs where bgs.obi=obiv; 

commit; 

select count(*) into bgs_chk from bgs bgs; 

If(bgs_chk = 0) then 

Key1V := Key1V + 1; 

Key2V :=1; 

else 

select obi into obiv from bgs bgs where bgs.ibi=ibiv; 

goto label2; 

end if; 

end if; 

end loop; 

END ; 

/ 

table script 

create table bio 

( 

obi varchar2(255), 

ibi varchar2(255), 

load_date date, 

update_date date) 

create table oo( 

sort_num number, 





create table oi( 


oi number, 

output_count number, 



create table bgs( 



seq_num number, 



create table bgr( 

batch_num varchar2(255), 

key_1 number, 

key_2 number, 

gen_sequence number, 



Thanks in advance. 


... sometimes sorting fails. .. 

NO, it does not. You might have coded something incorrectly, but "sometimes sorting fails" is not an accurate statement to 

make.

make. 

but I would like you to read this posting you made, and - pretend you didn't know the problem attempting to be solved at all 

(sort of like us for example) and ask yourself "would anyone really be able to tell what I was trying to do given what I've 

posted - did I give them a clear definition of the problem in written words" 

the answer to that is "absolutely not, there is no problem definition, there is however a lot of code that is known to not 

function correctly from which one could learn how not to solve the as yet to be stated problem".... 

Great February 16, 2007 - 8pm Central time zone Bookmark | Bottom | Top 

Reviewer: Reader from BOSTON 

Tom 

You should take off from work and provide the answer :) 

Different Result March 13, 2007 - 4pm Central time zone Bookmark | Bottom | Top 

Reviewer: Chi from Peekskill, NY 

Tom: 

I ran the following SQL on my development & production servers and got different results: 

select 1 from plateau.pa_student a 

where a.stud_id = 'adhh01' 

and a.stud_id in 

(select b.stud_id 

from plateau.pa_student b start with b.stud_id = 'deaalder' connect by prior b.super = 

b.stud_id 

union select c.stud_id 

from plateau.pa_student c start with c.stud_id = 'deaalder' connect by c.super = prior 

c.stud_id ) 

Both servers are running 10.2.0.3 on Windows 2003 and all instances are created using DBCA. On the development server 

the result was as expected: "no rows selected". However on the production server it returned "1". I have compared all 

parameters on these 2 instances and found no mismatch (except those include the path). The problem is with the server 

as I tested on another instance on the production server and got same incorrest result. What would cause this kind of 

discrepancy? Thanks for the help. 


perhaps different data... 

but, you give us nothing to work with. 

are the plans the same? 

Display all dates in a year March 19, 2007 - 3pm Central time zone Bookmark | Bottom | Top 

Reviewer: Jay from Herndon, VA 

Good day Tom! 

I honestly am hoping that I am not posting this question in a wrong thread. 

Tom, I am trying to just write a query to give me all the dates from a particular date to the system date. 

For instance, if I enter 01/01/2007, I need a query to fetch me records like this... 

01/01/2007 

01/02/2007 

01/03/2007 

01/04/2007 

... 

.. 

03/19/2007 

Is this something simple to do? Do we need to take into consideration the leap year issue as well or will oracle know the 

dates? 

Thanks Tom. Have a good one!


ops$tkyte%ORA10GR2> with data 

2 as 

3 (select level-1 l 

4 from dual 

5 connect by level 



Tom, 

Can you explain the logic behind 

'connect by level < n' when the from clause has Dual or a Single Row rowsource. 

Looks like the connect by keeps pumping rows because it cannot decide a relationship between parent and child rows. 

Is it like if it can't find a relationship, it will assume each row to be a parent as well as a child for every level? 

2) Why did you not choose to do it to a multi row table? 


Ravi 


with data 

as 

(select level from dual connect by level < :n ) 

is just a way to create :N rows - the connect by statement is satisfied as long as level is less than N - so it gets the first row 

from dual, sets level to 1 and says "1 is less than :n, great, we therefore create a new level and try again, now level is 2....." 

2) because I needed N rows and dual is perfect for that? If I picked a 'real table', I'd have to pick on with at least :n rows - but 

we don't know what :n is 

wow! March 20, 2007 - 8am Central time zone Bookmark | Bottom | Top 

Reviewer: Jay from Herndon, VA 

Thanks Tom! 

Jay 

Kindly help SQL for Chain Marketing system - March 31, 2007 - 10am Central time zone 


Reviewer: Arindam Mukherjee from Kolkata, India 

Chain Marketing system, one has so many workers. That worker also has so many workers. In this way, we get so many 

workers down the line forming a chain. Our table structure looks like below. The total no. of records in this table is equal to 

the no. of agents. So Agent_id and Introducer_id are self_referential. 

Agent_id | Commission | GAP-Commi| Introducer_id | Introducer_rank | team_strength 

Data looks like

1st row >> 100 | 15% | 0 | 91 | 2 | 0 

2nd row >> 91 | 5% | 0 | 81 | 3 | 7 

3rd row >> 81 | 2% | 3% |61 | 5 | 9 --- here 4th rank is missing because of termination. 

31st row >> 300 | 15% | 0 | 51 | 2 | 0 

32nd row >> 51 | 5% | 0 | 47 | 3 | 18 

33rd row >> 47 | 1% | 3% + 2% |33 | 6 | 56 --- here 4th and 5th rank is missing because of termination. 

We need the following results. 

1 > Each rank and its strength down the line, suppose for agent ID = 81, we need 

For 81, 9 and for 91, 7 as 91 is related to 81. 

2> When one agent brings one business say $100, commission column will be updated as follows. 

1st row >> 15% of 100 = 15 (Either insert or update with existing one) 

2nd row >> 5% of 100 = 5 (Either insert or update with existing one) 

3rd row >> 2% of 100 = 2 (Either insert or update with existing one) 

3rd row >> Gap Commission 3% of 100 = 3 (Either insert or update with existing one) 

Since 4th Rank does not exist, 5th rank will get Gap commission. 

You have every right to change the table structure but please help us write these complicated SQL as the table will have 

more than 40 thousands record. We are trying " connect by " clause but could not get success. Our database is Oracle 9i. 

Kindly help us. 

April 1, 2007 - 11pm Central time zone Bookmark | Bottom | Top 

Reviewer: Arindam Mukherjee from Kolkata, India 

Sir, 

You please read the above facts and kindly guide me how to calculate commission when one business of $100 would be 

in place. So I need Insert / Update and Query SQL, Please help me. 

Performance of connect by April 13, 2007 - 8am Central time zone Bookmark | Bottom | Top 

Reviewer: Artur Popov from Ukhta, Komi, Russian Federation 

Hi, Tom. 

I have a big problem with two queries and I want to understand why do their execution time differs so much. Here they are: 

SELECT ap.app_id, ap.created, u.fio_short from_user, ut.fio_short to_user, ty.TYPE_NAME, ap.app_num 

FROM applications_events ae, apps ap, application_types_ref ty, isa_own.users u, isa_own.users ut 

WHERE ae.application_id = ap.app_id 

AND ae.event_time = ap.recent_event_time 

AND ae.event_id = 3 

AND ae.user_id IN 

(SELECT id_user 

FROM isa_own.users 

WHERE pr_rabot = 1 START WITH id_user = 363 CONNECT BY PRIOR id_user = id_nach) 

AND u.id_user = ap.creator_id 

AND ut.id_user = ae.user_id 

AND ap.TYPE_ID = ty.TYPE_ID; 

It takes over 5 seconds to execute. But when I replaced the subquery with connect by with it's result: 

SELECT ap.app_id, ap.created, u.fio_short from_user, ut.fio_short to_user, ty.TYPE_NAME, ap.app_num 

FROM applications_events ae, apps ap, application_types_ref ty, isa_own.users u, isa_own.users ut 

WHERE ae.application_id = ap.app_id 

AND ae.event_time = ap.recent_event_time 

AND ae.event_id = 3 

AND ae.user_id IN 

(363,359,361,364,341,354,590,591,944,840) 

AND u.id_user = ap.creator_id 

AND ut.id_user = ae.user_id 

AND ap.TYPE_ID = ty.TYPE_ID

the execution time became just 0.9s. 

Why does it happens and how can I speed up my first query? 

Followup April 13, 2007 - 2pm Central time zone: 

did you look at the plans to see what is fundamentally different 

and how long does the connect by take itself. 


Reviewer: Popov Artur from Ukhta, Komi, Russian Federation 

Hi Tom. 

Here are the plans for my queries from the previous question. 

First query: 


------- --------------------------------------------------- 

128 HASH JOIN (cr=6551 r=5769 w=1529 time=5676858 us) 



128 MERGE JOIN (cr=6480 r=5769 w=1529 time=5662939 us) 

4032 SORT JOIN (cr=6428 r=5769 w=1529 time=5615042 us) 

4032 MERGE JOIN SEMI (cr=6428 r=5769 w=1529 time=5586127 us) 


384128 TABLE ACCESS FULL APPLICATIONS_EVENTS (cr=6363 r=4240 w=0 time=907248 us) 

4032 SORT UNIQUE (cr=65 r=0 w=0 time=609225 us) 

10 VIEW (cr=65 r=0 w=0 time=13402 us) 

10 FILTER (cr=65 r=0 w=0 time=13379 us) 

11 CONNECT BY WITH FILTERING (cr=65 r=0 w=0 time=13344 us) 

1 NESTED LOOPS (cr=3 r=0 w=0 time=133 us) 

1 INDEX UNIQUE SCAN PK_ID_USER (cr=2 r=0 w=0 time=72 us)(object id 10236) 

1 TABLE ACCESS BY USER ROWID USERS (cr=1 r=0 w=0 time=49 us) 


11 CONNECT BY PUMP (cr=0 r=0 w=0 time=92 us) 

1920 TABLE ACCESS FULL USERS (cr=62 r=0 w=0 time=4132 us) 


6000 TABLE ACCESS FULL APPS (cr=52 r=0 w=0 time=9611 us) 

2 TABLE ACCESS FULL APPLICATION_TYPES_REF (cr=7 r=0 w=0 time=190 us) 



Second query: 


------- --------------------------------------------------- 




4032 TABLE ACCESS BY INDEX ROWID APPLICATIONS_EVENTS (cr=2125 r=28 w=0 time=78225 us) 


10 INLIST ITERATOR (cr=31 r=0 w=0 time=546 us) 

10 TABLE ACCESS BY INDEX ROWID USERS (cr=31 r=0 w=0 time=416 us) 

10 INDEX RANGE SCAN PK_ID_USER (cr=21 r=0 w=0 time=265 us)(object id 10236) 

4352 INDEX RANGE SCAN T_INDEX3 (cr=51 r=0 w=0 time=16801 us)(object id 29522) 

6000 TABLE ACCESS FULL APPS (cr=54 r=0 w=0 time=8504 us) 

128 TABLE ACCESS BY INDEX ROWID APPLICATION_TYPES_REF (cr=138 r=0 w=0 time=2066 us) 

128 INDEX UNIQUE SCAN PK_APPLICATIONS_TYPES_REF (cr=10 r=0 w=0 time=751 us)(object id 9935) 

128 TABLE ACCESS BY INDEX ROWID USERS (cr=266 r=0 w=0 time=1995 us) 

128 INDEX UNIQUE SCAN PK_ID_USER (cr=138 r=0 w=0 time=942 us)(object id 10236) 

We can see, that they are completely different, but I didn't change the query, just replaced the subquery with it's result. Why 

did it cause such consequences? 


so, in the bad plan - are the estimated card= values near to the actuals you posted here. 

NO_FILTER Hint - What is it ? April 16, 2007 - 12am Central time zone Bookmark | Bottom | Top

NO_FILTER Hint - What is it ? April 16, 2007 - 12am Central time zone Bookmark | Bottom | Top 

Reviewer: BC from Macomb Twp 

Tom, 

What is the NO_FILTER hint ? I have seen it used in several queries using CONNECT BY. 


BC 



Statistics for all tables is gathered and actual. 

For the first query I got this plan (using autotrace): 

0 SELECT STATEMENT Optimizer=CHOOSE (Cost=514 Card=16 Bytes=2656) 

1 0 HASH JOIN (Cost=514 Card=16 Bytes=2656) 



4 3 MERGE JOIN (Cost=501 Card=16 Bytes=1328) 

5 4 SORT (JOIN) (Cost=435 Card=192224 Bytes=6727840) 

6 5 MERGE JOIN (SEMI) (Cost=435 Card=192224 Bytes=6727840) 


8 7 TABLE ACCESS (FULL) OF 'T' (Cost=427 Card=192224 Bytes=4228928) 

9 6 SORT (UNIQUE) (Cost=8 Card=960 Bytes=12480) 

10 9 VIEW OF 'VW_NSO_1' (Cost=4 Card=960 Bytes=12480) 

11 10 FILTER 



14 13 INDEX (UNIQUE SCAN) OF 'PK_ID_USER' (UNIQUE) (Cost=1 Card=1 

Bytes=4) 

15 13 TABLE ACCESS (BY USER ROWID) OF 'USERS' 

16 12 HASH JOIN 


18 16 TABLE ACCESS (FULL) OF 'USERS' (Cost=4 Card=960 Bytes=10560) 


20 19 TABLE ACCESS (FULL) OF 'APPS' (Cost=7 Card=6000 Bytes=288000) 

21 3 TABLE ACCESS (FULL) OF 'APPLICATION_TYPES_REF' (Cost=2 Card=2 Bytes=94) 



Ok, to make everything clear here is my tables: 

-- 6000 rows. 

create table APPS 

( 

APP_ID NUMBER, 

TYPE_ID NUMBER not null, 

CREATOR_ID NUMBER(38) not null, 

CREATED TIMESTAMP(6) not null, 

POSSIBLE_NAPR_ID NUMBER not null, 

RECENT_EVENT_TIME TIMESTAMP(4), 

APP_NUM VARCHAR2(14) 

); 

alter table APPS add constraint PK_APPS primary key (APP_ID); 

alter table APPS add constraint FK_APPS_NAPR foreign key (POSSIBLE_NAPR_ID) 

references ISA_OWN.NAPRAVL (ID_NAPR); 

alter table APPS add constraint FK_APPS_TYPES foreign key (TYPE_ID) 

references APPLICATION_TYPES_REF (TYPE_ID); 

alter table APPS add constraint FK_APPS_USERS foreign key (CREATOR_ID) 

references ISA_OWN.USERS (ID_USER); 

-- The following table is made just for testing. 

-- It have 768383 rows. 

create table APPLICATIONS_EVENTS 

( 

EVENT_TIME TIMESTAMP(4) not null, 

EVENT_ID NUMBER not null, 

APPLICATION_ID NUMBER not null, 

USER_ID NUMBER, 

REASON VARCHAR2(50), 

NAPR_ID NUMBER,

NAPR_ID NUMBER, 

ITEM_ID NUMBER 

) 

create index T_INDEX1 on APPLICATIONS_EVENTS (EVENT_TIME); 

create index T_INDEX2 on APPLICATIONS_EVENTS (APPLICATION_ID); 

create index T_INDEX3 on APPLICATIONS_EVENTS (USER_ID); 

-- 2 rows at this moment. 

create table APPLICATION_TYPES_REF 

( 

TYPE_ID NUMBER not null, 

TYPE_NAME VARCHAR2(60) not null 

); 

alter table APPLICATION_TYPES_REF add constraint PK_APPLICATIONS_TYPES_REF primary key (TYPE_ID); 

-- 920 rows. 

create table ISA_OWN.USERS 

( 

FNAME VARCHAR2(50) not null, 

MNAME VARCHAR2(50) not null, 

LNAME VARCHAR2(50) not null, 

FIO_SHORT VARCHAR2(60) not null, 

ID_USER NUMBER(10) not null, 

FOTO BLOB, 

KAB VARCHAR2(50), 

TEL VARCHAR2(50), 

EMAIL VARCHAR2(50), 

ID_DOLJNOST NUMBER(10) not null, 

ID_NAPR NUMBER(10) not null, 

TAB VARCHAR2(5) not null, 

FOTO_NAME VARCHAR2(100), 

CON_NAME VARCHAR2(50) not null, 

PAS VARCHAR2(50) not null, 

VID_DOG VARCHAR2(20), 

DATE_NAIM DATE, 

ID_NACH NUMBER(10), 

PR_RABOT NUMBER(1) default 1 not null, 

D_R DATE, 

TEL_DOM VARCHAR2(50), 

PROFSOUZ NUMBER(1) default 1 not null, 

POL VARCHAR2(1) not null, 

EMAIL_LOC VARCHAR2(50), 

NOVELL_NAME VARCHAR2(20) 

); 

alter table ISA_OWN.USERS add constraint PK_ID_USER primary key (ID_USER); 

alter table ISA_OWN.USERS add constraint CON_PAS_UNIQ unique (CON_NAME,PAS); 

alter table ISA_OWN.USERS add constraint TAB_UNIQ unique (TAB); 

create index ISA_OWN.IND_ID_NACH on ISA_OWN.USERS (ID_NACH); 

My colleagues adviced me to store the results of the 'connect by' subquery in a temporary table and replace the 'connect by' 

subquery with simple select from this table, but I think this is not the best idea. What can you say? 

Followup April 17, 2007 - 9am Central time zone: 

I simply asked: 

so, in the bad plan - are the estimated card= values near to the actuals you posted here. 



As you can see they are close, excluding APPLICATIONS_EVENTS. I just wanted to give you better understanding of the 

problem. 

So, can you help? 

Connect by prior July 18, 2007 - 11am Central time zone Bookmark | Bottom | Top 

Reviewer: Bob from London, UK 

Hi Tom, 

Is there a way of listing parent and child relations in a query. For example, if Site 1: has child sites 2,3,4,5,6. 

I want to able to select: 

1,2

1,2 

1,3 

1,4 

1,5 

1,6 

and so on.. 



from what do you want to select this 

connect by prior July 18, 2007 - 11am Central time zone Bookmark | Bottom | Top 

Reviewer: Bob from London 

Good explanation at the top, answers my question!!! 

Connect by mecanism June 23, 2008 - 1am Central time zone Bookmark | Bottom | Top 

Reviewer: Car Elcaro 

I see the mecanism of connect by and join above in this url : 

 

http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:489772591421#69739221126046 

join is done 

start with applied 

connect by done 

where clause applied to results 

I have confusion here : 

create table t1 

( 

key number, 

parent varchar2(2) 

) 


( 

parent varchar2(2), 

child varchar2(2) 

) 

insert into t1 values (1,'A1'); 

insert into t1 values (2,'B1'); 

select * from t1; 

KEY PA 

---------- -- 

1 A1 

2 B1 

insert into t2 values ('A1','A2'); 




PA CH 

-- -- 

A1 A2 

A2 A3 

A3 A4 

Join between t1 and t2 without any condition 

select t1.key, t1.parent, t2.child from t1, t2;

KEY PA CH 

---------- -- -- 

1 A1 A2 

2 B1 A2 

1 A1 A3 

2 B1 A3 

1 A1 A4 

2 B1 A4 


Use connect by with 'start with' clause 

column scbp format a20 

select level, key, 

sys_connect_by_path(key || ',' || child ,'|') scbp 

from t1, t2 

start with t2.parent = t1.parent 

connect by prior t2.child = t2.parent 

order by level 

LEVEL KEY SCBP 

---------- ---------- -------------------- 

1 1 |1,A2 

2 2 |1,A2|2,A3 

2 1 |1,A2|1,A3 

3 1 |1,A2|2,A3|1,A4 

3 2 |1,A2|2,A3|2,A4 

3 1 |1,A2|1,A3|1,A4 

3 2 |1,A2|1,A3|2,A4 


Here I only see one level 1, why ? My reason asking here is that you said join first and then 'start with' clause executed and 

from intermediate join result between t1 and t2 with parent = 'A1' is 3 record. 

select * from 

( 

select t1.key, t1.parent, t2.child from t1, t2 

) 

where parent = 'A1' 

KEY PA CH 

---------- -- -- 

1 A1 A2 

1 A1 A3 

1 A1 A4 

Questions : 

1. Why Oracle not choose A3 as a level one (see SCBP result above) ? Does it just lucky ? Is Oracle implicitly add predicate 

key = 1 as parent = 'A1' is originated from key = 1 from table t1 ? 

2. Why only 1 level one not three ? 

3. Query to show output like below - of course without pl/sql : 

--insert first at table t2 

insert into t2 values ('B1','B2'); 



KEY ID 

---------- ---------- 

1 A2 

1 A3 

1 A4 

2 B2 

2 B3 

2 B4

Thanks. 


You queried: 

select t1.key, t1.parent, t2.child from t1, t2; 

and then you 

select level, key, 

sys_connect_by_path(key || ',' || child ,'|') scbp 

from t1, t2 

start with t2.parent = t1.parent 

connect by prior t2.child = t2.parent 

order by level 

used parent from both (but only queried parent from one) and used t2.parent again, but didn't query it. So, I fail to see why 

the cartesian join you presented is relevant - it isn't showing most of the data you use. 

ops$tkyte%ORA10GR2> create table t3 

2 as 

3 select t1.key t1_key, t1.parent t1_parent, t2.child t2_child, t2.parent t2_parent from t1, t2; 


ops$tkyte%ORA10GR2> 

ops$tkyte%ORA10GR2> select * from t3; 

T1_KEY T1_PARENT T2_CHILD T2_PARENT 

---------- --------- -------- --------- 

1 A1 A2 A1 

1 A1 A3 A2 

1 A1 A4 A3 

2 B1 A2 A1 

2 B1 A3 A2 

2 B1 A4 A3 



ops$tkyte%ORA10GR2> column scbp format a20 


ops$tkyte%ORA10GR2> select level, key, 

2 sys_connect_by_path(key || ',' || child ,'|') scbp 

3 from t1, t2 

4 start with t2.parent = t1.parent 

5 connect by prior t2.child = t2.parent 

6 order by level; 

LEVEL KEY SCBP 

---------- ---------- -------------------- 

1 1 |1,A2 

2 2 |1,A2|2,A3 

2 1 |1,A2|1,A3 

3 1 |1,A2|2,A3|1,A4 

3 2 |1,A2|2,A3|2,A4 

3 1 |1,A2|1,A3|1,A4 

3 2 |1,A2|1,A3|2,A4 


so, start with set is: 

ops$tkyte%ORA10GR2> select * from t3 where t2_parent = t1_parent; 

T1_KEY T1_PARENT T2_CHILD T2_PARENT

---------- --------- -------- --------- 

1 A1 A2 A1 

a single row to start with... and then the first connect by level (you said there is only one level, but your example shows 

THREE!!!!) would be: 

ops$tkyte%ORA10GR2> select * from t3 where t2_parent = 'A2'; 


---------- --------- -------- --------- 

1 A1 A3 A2 

2 B1 A3 A2 

and then level three would be: 



---------- --------- -------- --------- 

1 A1 A4 A3 

2 B1 A4 A3 

and level four: 



level 1 connected to 2 level 2's. Each of the level 2's connected to 2 level 3's. Level 3 was the end. 

ops$tkyte%ORA10GR2> select rpad('*', 2*level, '*') indent, level, key, 

2 sys_connect_by_path(key || ',' || child ,'|') scbp 

3 from t1, t2 

4 start with t2.parent = t1.parent 

5 connect by prior t2.child = t2.parent 

6 / 

INDENT LEVEL KEY SCBP 

---------- ---------- ---------- -------------------- 

** 1 1 |1,A2 

**** 2 1 |1,A2|1,A3 

****** 3 1 |1,A2|1,A3|1,A4 

****** 3 2 |1,A2|1,A3|2,A4 

**** 2 2 |1,A2|2,A3 

****** 3 1 |1,A2|2,A3|1,A4 

****** 3 2 |1,A2|2,A3|2,A4 


Now, your questions don't make sense because they were based on a flawed example (it is easy to see why there is only 

one level one once you look at all of the columns, there isn't one level, there are THREE and your example shows that) 

and the kicker question, #3, well, please. Tell us the logic there. I have no idea what your output represents and given that 

the example was not very useful - we are really lost in the weeds.

e: Connect by mecanism June 25, 2008 - 1am Central time zone Bookmark | Bottom | Top 


assume that t2 is a process table (t2.child) and and t1.key is a row material. both table are 

connected with t1.parent and t2.parent as a starting point and then we just want to see all process 

in t2.child that belongs to t1.id. 

Here is the example 

t1.key = 1 t1.parent = 'A1' we get from t2 as starting point is (A1,A2) and then connected to it 

self so resulting A2, A3, A4 as processes that key = 1 is pass. 


this is no example 

assume we don't know what t1 looks like (therefore, need a create) 

assume same for t2 

assume we haven't any data in either of them and need some. 

assume "row material" is a meaningless term - I've never heard it. 

assume that t1.id and t2.child have some relationship we don't understand "then we just want to see all process in t2.child 

that belongs to t1.id." is not meaningful to us. 

Build better example please - this one was not intuitive. Don't say "i already gave you tables", you did, but not for this 

example. t1.id is a new thing here. 

ignore the existence of the prior example, make everything self contained and clear with true examples. Perhaps build it up 

the way I did to demonstrate how connect by works above - a bit a time, building layer upon layer showing you what I meant. 

completing question on connect by June 28, 2008 - 1am Central time zone Bookmark | Bottom | Top 


 

Sorry for incomplete and inconsitent question above. Let me reintroduce the problem. 

I have table t2 defined as follow : 


( c_inp varchar2(2), 

c_out varchar2(2), 

c_prc varchar2(2) 

); 

insert into t2 values ('R1','S1','P1'); 

insert into t2 values ('S1','T1','P2'); 

insert into t2 values ('T1','U1','P3'); 

insert into t2 values ('U1','V1','P4'); 

insert into t2 values ('R2','S2','P1'); 

insert into t2 values ('S2','T2','P2'); 

insert into t2 values ('T2','U2','P3'); 

insert into t2 values ('U2','V2','P4'); 

insert into t2 values ('V2','W2','P5'); 

commit; 

column c_inp format a5; 

column c_out format a5; 

column c_prc format a5; 


C_INP C_OUT C_PRC 

----- ----- ----- 

R1 S1 P1 

S1 T1 P2 

T1 U1 P3 

U1 V1 P4 

R2 S2 P1 

S2 T2 P2

S2 T2 P2 

T2 U2 P3 

U2 V2 P4 

V2 W2 P5 


c_inp is RAW MATERIAL or INTERMEDIATE MATERIAL to process specified in c_prc into c_out so each row of this table 

mean a process. For example R1 is a RAW MATERIAL processed by P1 into S1. Next S1 is an INTERMEDIATE MATERIAL 

to process into T1. Similar to say that S1,T1,U1,V1 of c_inp is INTERMEDIATE OUTPUT for R1,S1,T1,U1 of c_out 

respectively. 

I have another table that contain who is the supplier of raw material like R1 and R2 on the table above. Here is the structure 

: 


( 

c_sup varchar2(4), 

c_raw varchar2(2) 

); 

insert into t1 values ('SUP1','R1'); 

insert into t1 values ('SUP2','R2'); 

commit; 

column c_sup format a5; 

column c_raw format a5; 


C_SUP C_RAW 

----- ----- 

SUP1 R1 

SUP2 R2 

Actually question #3 : I want to list supplier with their INTERMEDIATE OUTPUT so for supplier SUP1 I have four intermediate 

output (S1,T1,U1,V1) and for SUP2 I have five (S2,T2,U2,V2,W2). Expected result : 

C_SUP C_OUT 

----- ----- 

SUP1 S1 

SUP1 T1 

SUP1 U1 

SUP1 V1 

SUP2 S2 

SUP2 T2 

SUP2 U2 

SUP2 V2 

SUP2 W2 


Currently, my solution use a PL/SQL to solve it. And I search a better solution if it's exist e.g. without using PL/SQL. Here is 

my current solution : 

create or replace function find_output(p_raw varchar2) 

return varchar2 

is 

l_output varchar2(20); 

begin 

select max(sys_connect_by_path(c_out,',') || ',') 

into l_output 

from t2 

start with c_inp = p_raw 

connect by c_inp = prior c_out; 

return l_output;

eturn l_output; 

end; 

/ 

select c_sup, column_value c_out 

from (select c_sup, find_output(c_raw) c_list from t1) a, 

table( 

cast( 

multiset( 

(select substr(c_list,instr(c_list,',',1,level)+1, 

instr(c_list,',',1,level+1)instr(c_list,',',1,level)-1) 

from dual 

connect by level

more efficient. 

To : Car Elcaro July 2, 2008 - 11am Central time zone Bookmark | Bottom | Top 

Reviewer: Raj from United Kingdom 

Tom, 

First and foremost I thank you for your wonderful support to the oracle world. Correct if I am wrong. 

Why can't we do it this way. I assumed the OP is using oracle 10g or above. 

SQL> select * from v$version; 

BANNER 

---------------------------------------------------------------- 

Oracle Database 10g Enterprise Edition Release 10.2.0.2.0 - 64bi 

PL/SQL Release 10.2.0.2.0 - Production 

CORE 10.2.0.2.0 Production 

TNS for HPUX: Version 10.2.0.2.0 - Production 

NLSRTL Version 10.2.0.2.0 - Production 

SQL> select * from t1; 

C_SUP C_RAW 

---------- ---------- 

SUP1 R1 

SUP2 R2 

SQL> select * from t2; 

C_INP C_OUT C_PRC 

---------- ---------- ---------- 

R1 S1 P1 

S1 T1 P2 

T1 U1 P3 

U1 V1 P4 

R2 S2 P1 

S2 T2 P2 

T2 U2 P3 

U2 V2 P4 

V2 W2 P5 


SQL> l 

1 select c_sup, c_out 

2 from 

3 ( 

4 select connect_by_root c_inp root_val, c_out from t2 

5 start with c_inp in (select c_raw from t1) 

6 connect by prior c_out = c_inp 

7 )t, t1 

8* where t.root_val = t1.c_raw 

SQL> / 

C_SUP C_OUT 

---------- ---------- 

SUP1 V1 

SUP1 U1 

SUP1 T1 

SUP1 S1 

SUP2 V2 

SUP2 U2 

SUP2 T2 

SUP2 W2 

SUP2 S2 


Explain plan for the same. 

SQL> / 

PLAN_TABLE_OUTPUT 

----------------------------------------------------------------------------------------------------


------------------------------------------------------------------------------------ 


------------------------------------------------------------------------------------ 


|* 1 | HASH JOIN | | 9 | 126 | 7 (15)| 00:00:01 | 

| 2 | TABLE ACCESS FULL | T1 | 2 | 16 | 3 (0)| 00:00:01 | 

| 3 | VIEW | | 9 | 54 | 3 (0)| 00:00:01 | 

|* 4 | CONNECT BY WITH FILTERING| | | | | | 

|* 5 | FILTER | | | | | | 


|* 7 | TABLE ACCESS FULL | T1 | 1 | 3 | 3 (0)| 00:00:01 | 

|* 8 | HASH JOIN | | | | | | 




------------------------------------------------------------------------------------ 


--------------------------------------------------- 

1 - access("T"."ROOT_VAL"="T1"."C_RAW") 

4 - access("C_INP"=PRIOR "C_OUT") 

5 - filter( EXISTS (SELECT 0 FROM "T1" "T1" WHERE "C_RAW"=:B1)) 

7 - filter("C_RAW"=:B1) 

8 - access("C_INP"=PRIOR "C_OUT") 


Regards 

Raj 


big page 

no idea who "OP" is - the original poster asked about "how does connect by work". 

Remember - there are an infinite number of answers to pretty much every question. So the answer in general to "why can't 

we do it this way" is - well, of course you can - you can do something like this, that or the other way. 

Not that I verified your approach - I didn't really piece together what you were trying to solve - just saying "there are 

thousands of ways to do anything" 

July 7, 2008 - 9am Central time zone Bookmark | Bottom | Top 

Reviewer: Raj from United Kingdom 

Tom, 

I thought since I have mentioned (To : Car Elcaro) name in the subject of my post I assumed it 

would have correlated my reference of OP to be this reviewer, apparently it didn't. So to be more 

precise I posted one of the solutions which could be used for his question "Currently, my solution 

use a PL/SQL to solve it. And I search a better solution if it's exist e.g. without using PL/SQL." 

Hope it makes sense this time. 

Regards 

Raj 

using connect by to replace self join August 5, 2008 - 10pm Central time zone 

Reviewer: A reader from NJ 

I am referring to your post (reproduced below) on this page earlier where you suggested to use a 

self join: 

(Dated May 7, 2006 - 11am US/Eastern:) 

Bookmark | Bottom | Top

"query all Employees ( lets say whose sal >=3000 ) and their managers." 

why would you use connect by? 

ops$tkyte@ORA10GR2> select ename, mgr, sal from emp where sal >= 3000; 

ENAME MGR SAL 

---------- ---------- ---------- 

SCOTT 7566 3000 

KING 5000 

FORD 7566 3000 

or if you needed their manager name: 

1 select a.ename, b.ename mgr, a.sal 

2 from emp a, emp b 

3 where a.sal >= 3000 

4* and a.mgr = b.empno(+) 

ops$tkyte@ORA10GR2> / 

ENAME MGR SAL 

---------- ---------- ---------- 

FORD JONES 3000 

SCOTT JONES 3000 

KING 5000 

My Question: 

(ORACLE VERSION: 10.2.0.2) 

I have a view that does a self join to a table that is 25 million rows. Very similar to what you 

suggested above. The self join's only purpose is to get employee's manager name. But the self join 

results in nested loop outer, causing very high lio. Query on this view is executed hundreds of 

times in a sec, causing CPU related issues. 

I just want to see if connect by will give me a better execution plan, but I do not know how to 

write a connect by which will give me this: 

empno ename job manager's job 

7839 KING PRESIDENT NULL 

7566 JONES MANAGER PRESIDENT 

7788 SCOTT ANALYST MANAGER 

7369 SMITH CLERK ANALYST 

.. 

Can you please help me write one? I tried sys_connect_by_path, but could not go further. 


are you using the cbo or rbo and what are the estimated cardinalities in the plan. 

the optimizer would not use nested loops unless you did something like hint it or used first rows optimization. 

show us the plan and explain your optimizer environment. 

but wait, if you have a query that returns 25 million rows and is executed "hundreds of times per second", then you must not 

be returning 25 million rows, you must be returning a very very very very small subset 

so, you need to give us more information. 

and lose the connect by idea, that doesn't apply, doesn't make sense. 

February 12, 2009 - 11pm Central time zone Bookmark | Bottom | Top 


Tom, 

Need to take care of symbol changes in our database and calculate the average volume for last 10 

days. 

create table pricing 

(ticker varchar2(10), 

px_volume number, 

date_loaded date);

create table ticker_changes 

(old_ticker varchar2(10), 

new_ticker varchar2(10), 

start_date date); 

insert into pricing 

values 

('aapl',100,to_date('08/18/2008','mm/dd/yyyy')); 


values 



values 



values 


commit; 

on 08/22/2008, aapl changes to abc and this is inserted to ticker_changes table. From 08/22/2008 

onwards, pricing table has abc instead of aapl 

insert into ticker_changes 

values 

('aapl','abc',to_data('08/22/2008','mm/dd/yyyy')); 

My output should be as shown below for 08/21/2008 

dt ticker avg10d 

08/21/2008 aapl 100 

10 days before 08/21/2008 is 08/11/2008. I need to take the px_volume for all these days and divide 

by 10. But I only have data on 08/18,08/19,08/20,08/21. Regardless of if I have data on any of 

these 10 days, I have to take the sum of the px_volume and divide by 10. So, my average is 

(100+200+300+400)/20=100 for 08/21/2008 

On 08/22/2008, there is a change to ticker = aapl. aapl has been changed to abc. In pricing table, 

I will insert one record on 08/22/2008 


values 

('abc',500,to_date('08/22/2008','mm/dd/yyyy')); 


08/22/2008 abc 150 


by 10. But I only have data on 08/18,08/19,08/20,08/21,08/22 Regardless of if I have data on any of 

these 10 days, I have to take the sum of the px_volume and divide by 10. So, my average is 

(100+200+300+400+500)/20=150 for 08/22/2008. Here for my calculations I need to consider aapl from 

08/12/2008 to 08/21/2008 and abc on 08/22/2008. I am assuming this can be done with hierachical 

queries. Can you guid me? 

On 08/25/2008 abc changes to pqr. I have a record in ticker_changes table 

insert into ticker_changes 

values 

('abc','pqr',to_data('08/25/2008','yyyy/mm/dd')); 


values 



values 



values 

('pqr',800,to_date('08/25/2008','mm/dd/yyyy'));


08/24/2008 abc 150 

08/25/2008 pqr 260 


by 10. 

(100+200+300+400+500+600+700+800)/20=260 for 08/25/2008. 

Can you tell me how to achieve this? 


read through this thread - almost identical 


6 


Reviewer: Reader 

Tom, 

I went through the other thread that you pointed out and wrote the query - 

select nvl(the_orig_sym,ticker) the_sym, 

date_loaded, 

px_volume 

from pricing p left outer join 

(select the_orig_sym, new_ticker, start_date sdate, 

nvl(lead(start_date) over (partition by the_orig_sym order by 

start_date)-1,to_date('31-DEC-9999','dd-mon-yyyy')) edate 

from (select connect_by_root old_ticker the_orig_sym, 

new_ticker, 

start_date 

from ticker_changes 

connect by prior new_ticker = old_ticker and prior start_date < start_date) 

) d 

on (p.date_loaded between d.sdate and d.edate and (p.ticker = d.new_ticker) ) 

where (sdate is null and edate is null) or (date_loaded between sdate and edate) 

order by /*ticker,*/ date_loaded; 

I get extra record when the start with connect by is used for the ticker_chages table, which is 

correct. When I try to join this table d with pricing, I am getting extra record for ticker=pqr on 

08/25/2008 as there are two records for pqr in the start with connect by query (table d in this 

case). When I run the query on 08/25, I need to get only one record. Can you tell me how to achieve 

this? 

this is the query I used to get the total of 10 days. 

select * 

from( 

select --nvl(the_orig_sym,ticker) the_sym, 

ticker, 

date_loaded, 

px_volume, 

sum(px_volume) over (partition by nvl(the_orig_sym,ticker) order by date_loaded desc rows 

between current row and 9 following) as sum_10d 

from pricing p left outer join 

(select the_orig_sym, new_ticker, start_date sdate, 




new_ticker, 

start_date 


connect by prior new_ticker = old_ticker and prior start_date < start_date) 

) d 

on (p.date_loaded between d.sdate and d.edate and (p.ticker = d.new_ticker) ) 

where (sdate is null and edate is null) or (date_loaded between sdate and edate) 

order by date_loaded ) 

where date_loaded= to_date('08/25/2008','mm/dd/yyyy')

TICKER DATE_LOADED PX_VOLUME SUM_10D 

---------- --------- ---------- ---------pqr 

25-AUG-08 800 800 

pqr 25-AUG-08 800 3600 

I should get 

TICKER DATE_LOADED PX_VOLUME SUM_10D 

---------- --------- ---------- ---------pqr 

25-AUG-08 800 3600 



Tom, 

I was wondering if you could answer the above question. 


first - how hard have you tried? do you understand how this works (if not, please get there before going further) 

second - make the test case really easy for me to follow - sort of a long narrative above - and you just say "I need one row, 

this one" right above - but don't really describe how you know "that is the row" (and perhaps when you do, what is needed 

will become obvious...) 

so, put it all together concisely (the test case to load the data - not scattered in various reviews, it is really hard for me to 

read up and down over more than one and figure out what is relevant, what is not - especially when hours/days and many 

other questions go in between your entries). But truly describe what needs to happen to the data, psuedo code it (like we 

used to in the olden days, to algorithmically describe it to someone...) even 


Tom, 

Thanks for your reply. 

February 18, 2009 - 12am Central time zone Bookmark | Bottom | Top 

I get data in pricing table every day. pricing table has data about each ticker traded in the 

market. For each ticker that comes in the pricing table, 10d sum has to be calculated and has to be 

inserted into other table. 

For example: 

--08/18 

insert into pricing values ('aapl',100,to_date('08/18/2008','mm/dd/yyyy')); 

insert into pricing values ('orcl',100,to_date('08/18/2008','mm/dd/yyyy')); 

--08/19 



--08/20 


insert into pricing values ('kkk',300,to_date('08/20/2008','mm/dd/yyyy')); 

--08/21 



on 08/18, if I run the query, the result is

select * 

from( 

select ticker, 

sum(px_volume) over (partition by ticker order by date_loaded desc rows between current 

row and 9 following) as sum_10d, 

date_loaded 

from pricing) 

where date_loaded = to_date('08/18/2008','mm/dd/yyyy') 

SQL> / 

TICKER SUM_10D DATE_LOAD 

---------- ---------- --------aapl 

100 18-AUG-08 

orcl 100 18-AUG-08 

On 08/19/2008, I get only aapl,orcl 

select * 

from( 




date_loaded 

from pricing) 


SQL> / 


---------- ---------- --------- 

aapl 300 19-AUG-08 

orcl 200 19-AUG-08 

Similary if I run the query on 08/20 

On 08/20/2008, if I run the query: 

select * 

from( 




date_loaded 

from pricing) 


SQL> / 


---------- ---------- --------aapl 

600 20-AUG-08 

kkk 300 20-AUG-08 

I have to run this for every day that I have data in pricing table 

Apparently, the ticker aapl got changed to abc on 08/22. And this data is inserted to 

ticker_changes table 

insert into ticker_changes values ('aapl','abc',to_data('08/22/2008','mm/dd/yyyy')); 

And in the pricing table, I will start getting abc instead of aapl. 

insert into pricing values ('abc',500,to_date('08/22/2008','mm/dd/yyyy')); 

When I run the query on 08/22 to get the 10d sum, I should, consider aapl, prior to 08/22 and abc 

from 08/22 onwards. 

The sum should be as shown below - 


---------- ---------- --------abc 

1500 20-AUG-08 

1500 is got from the below rows inserted to pricing table - 

aapl 100 18-AUG-08 

aapl 200 19-AUG-08 

aapl 300 20-AUG-08 

aapl 400 21-AUG-08 

abc 500 22-AUG-08

For this, I used the query that you pointed to in the other thread. 

select the_orig_sym, new_ticker, start_date sdate, 




new_ticker, 

start_date 


where start_date / 

THE_ORIG_S NEW_TICKER SDATE EDATE 

---------- ---------- --------- --------aapl 

abc 22-AUG-08 31-DEC-99 

On 08/22, when I get the sum for last 10 days, I have to add the following and I should get, 1500 

and it should be displayed under the ticker abc 

Ticker volume date_loaded 

aapl 100 18-AUG-08 

aapl 200 19-AUG-08 

aapl 300 20-AUG-08 

aapl 400 21-AUG-08 

abc 500 22-AUG-08 

Result for abc along with other tickers loaded on 08/22 

Ticker sum_10d date_loaded 

abc 1500 22-AUG-08 

(abc is actully the sum of aapl(prior to 08/22) and abc (on 08/22) 

--08/23 load to pricing table 


On 08/23, When I get the sum for last 10 days, I have to add the following and I should be getting, 

Ticker volume date_loaded 

aapl 100 18-AUG-08 

aapl 200 19-AUG-08 

aapl 300 20-AUG-08 

aapl 400 21-AUG-08 

abc 500 22-AUG-08 

abc 500 23-AUG-08 

Result for abc along with other tickers loaded on 08/23 

Ticker sum_10d DATE_LOADED 

abc 2000 08/23/2008 

Please let me know if I am missing something. 


please re-read my request above. 

... so, put it all together concisely ... not scattered in various reviews, it is really hard for me to read up and down over more 

than one and figure out what is relevant.... But truly describe what needs to happen to the data, psuedo code it (like we used 

to in the olden days, to algorithmically describe it to someone...) .... 

so, in looking at what you posted, would one have to go to more than one review section to get all of the necessary 

information? 

but that aside, if you have: 

select the_orig_sym, new_ticker, start_date sdate, 


start_date)-1,to_date('31-DEC-9999','dd-mon-yyyy')) edate



new_ticker, 

start_date 


where start_date / 

THE_ORIG_S NEW_TICKER SDATE EDATE 

---------- ---------- --------- --------aapl 

abc 22-AUG-08 31-DEC-99 

you basically have a mapping, the one you need, I see a single row. What is the issue, above you seemed to be saying "i 

get two" 

Connect By on DUAL table February 18, 2009 - 6am Central time zone Bookmark | Bottom | Top 

Reviewer: Matteo Mitrano from Italy 

Hi Tom, 

with regard to this present topic, could you please explain to me why I found this behaviour in my environment? Is there's 

something wrong with it? 

SQL> SELECT * FROM v$version; 

Oracle9i Release 9.2.0.1.0 - Production 

PL/SQL Release 9.2.0.1.0 - Production 

CORE 9.2.0.1.0 Production 

TNS for 32-bit Windows: Version 9.2.0.1.0 - Production 

NLSRTL Version 9.2.0.1.0 - Production 

Query 1: 

SELECT LEVEL - 1 AS set_of_days 

FROM DUAL 

CONNECT BY LEVEL


it was a 9i really old issue. 

you use an inline view as you did to avoid it. 



Tom, 

Regarding the previous question about the ticker changes: 

select mt.ticker 

,d.the_orig_sym 

,nvl(d.the_orig_sym,mt.ticker) the_sym 

,mt.date_loaded 

,mt.px_volume 

from (select * from pricing where date_loaded


where start_date

select mt.ticker 




,mt.px_volume 


ut I want to display only the record with the total and not pqr with px_volume=800 on 25th.Since 

pqr is present in pricing table on 25th, I need to get the 

sum for last 10 days, by considering ticker changes if any. 

if I run the below query, I get two records, one with the 10day sum 3600 and also one with the 

px_volume = 800 which was loaded on 25th 

select ticker,sum_10d,date_loaded 

from ( select mt.ticker 




,mt.px_volume 

,sum(px_volume) over (partition by nvl(the_orig_sym,ticker) order by date_loaded desc 

rows between current row and 9 following) as sum_10d 



new_ticker, 

start_date 


where start_date

How to acheive this? Hope I am clear enough in explaining this. Can I do a union between two 

hirearchical queries, one going down and other going up? If so, how can I go upwards to bring 

parent records? 

Thanks for all your help. 


... between two hirearchical queries, one going down and other going up? ... 

sure, but only because there is no such thing as "up or down", there are just links 

ops$tkyte%ORA10GR2> select rpad('*',2*level,'*') || ename nm, empno, mgr from scott.emp 



4 / 

NM EMPNO MGR 

--------------- ---------- ---------- 

**KING 7839 

****JONES 7566 7839 

******SCOTT 7788 7566 

********ADAMS 7876 7788 

******FORD 7902 7566 

********SMITH 7369 7902 

****BLAKE 7698 7839 

******ALLEN 7499 7698 

******WARD 7521 7698 

******MARTIN 7654 7698 

******TURNER 7844 7698 

******JAMES 7900 7698 

****CLARK 7782 7839 

******MILLER 7934 7782 



ops$tkyte%ORA10GR2> select rpad('*',2*level,'*') || ename nm, empno, mgr , 'down' 

2 from scott.emp 

3 start with ename = 'SCOTT' 


5 union all 

6 select rpad('*',2*level,'*') || ename nm, empno, mgr , 'up' 

7 from scott.emp 

8 start with ename = 'SCOTT' 

9 connect by prior mgr = empno 

10 / 

NM EMPNO MGR 'DOW 

--------------- ---------- ---------- ---- 

**SCOTT 7788 7566 down 

****ADAMS 7876 7788 down 

**SCOTT 7788 7566 up 

****JONES 7566 7839 up 

******KING 7839 up 



create table pricing 

(ticker varchar2(10), 

px_volume number, 

date_loaded date); 

create table ticker_changes 

(old_ticker varchar2(10), 

new_ticker varchar2(10), 

start_date date); 

--data for 08/18 

insert into pricing values ('aapl',100,to_date('08/18/2008','mm/dd/yyyy'));







insert into pricing values ('kkk',300,to_date('08/20/2008','mm/dd/yyyy')); 











insert into pricing values ('pqr',800,to_date('08/25/2008','mm/dd/yyyy')); 


--ticker changes insert 

insert into ticker_changes values ('aapl','abc',to_data('08/22/2008','mm/dd/yyyy')); 

insert into ticker_changes values ('abc','pqr',to_data('08/25/2008','yyyy/mm/dd')); 

commit; 

select * from ticker_changes order by start_date; 

OLD_TICKER NEW_TICKER START_DATE 

---------- ---------- ---------aapl 

abc 22-AUG-08 

abc pqr 25-AUG-08 

pqr aapl 26-AUG-08 

select * from pricing order by date_loaded; 

TICKER PX_VOLUME DATE_LOAD 

---------- ---------- --------orcl 

100 18-AUG-08 

aapl 100 18-AUG-08 

aapl 200 19-AUG-08 

orcl 100 19-AUG-08 

kkk 300 20-AUG-08 

aapl 300 20-AUG-08 

aapl 400 21-AUG-08 

orcl 100 22-AUG-08 

abc 500 22-AUG-08 

abc 600 23-AUG-08 

abc 700 24-AUG-08 

orcl 100 25-AUG-08 

pqr 800 25-AUG-08 

As shown in the above data from pricing table, I get tickers everday with there volume. There are 

chances that ticker can be changed. This ticker changes information is in ticker_changes table as 

shown above. 

Everyday, I need to use pricing table and caclulate last 10 day sum of volume (including that day 

in the 10 day) for each ticker in pricing table. I need to consider 

the ticker changes in the calculation. 

On 08/18/2008, I need to calculate sum for aapl, orcl. So the 10 day volume is from 08/08 to 08/18. 

Since I do not have data from 08/08 to 08/17, I use only the data from 08/18. 

TICKER TOTAL_VOLUME calculation_date

aapl 100 18-AUG-08 

orcl 100 18-AUG-08 

On 08/19, I have two tickers in pricing table, aapl and orcl. So the 10 day volume is from 08/08 to 

08/18. Since I have data from 08/18 and 08/19, I use that to calculate the volume. 

TICKER TOTAL_VOLUME calculation_date 

aapl 300 19-AUG-08 

orcl 200 19-AUG-08 

This continues everyday. On 08/22 as shown in the ticker_chages table "aapl" changed to "abc". For 

10 day volume calculation on 08/22, I need to consider aapl prior to 08/22 and abc on 08/22 


abc 1500 22-AUG-08 (this is sum of aapl+abc) 

orcl 300 22-AUG-08 

On 08/25, as shown in the ticker_changes table "abc" changed to "pqr", For 10 day volume 

calculation on 08/25, I need to consider aapl prior to 08/22, 

abc from 08/22 to 08/24 and pqr on 08/25 


abc 3600 22-AUG-08 (this is sum of aapl+abc) 

orcl 400 22-AUG-08 

I used the hierarchical query that you posted to get the ticker changes and the sdate and edate 

range. 

alter session set nls_date_format='DD-MON-YYYY'; 

select the_orig_sym, new_sym, effective_dt sdate, 

nvl(lead(effective_dt) over (partition by the_orig_sym order by 

effective_dt)-1,to_date('1-jan-3000','dd-mon-yyyy')) edate 

from ( select connect_by_root old_ticker the_orig_sym, new_ticker as new_sym, start_date as 

effective_dt 


where start_date


... If I use this to join to the pricing table, I get two records on 08/25 since I 

am joining on ticker_changes.new_sym with pricing.ticker 

..... 

easiest fix - just keep one of them - distinct or group by the set before applying the aggregates. 

Connect By issue April 3, 2009 - 3am Central time zone Bookmark | Bottom | Top 

Reviewer: Chandra from India 

Hi Tom, 

Please go through the below script. 

CREATE TABLE MATCHTABLE( 

MATCH1 VARCHAR2(10), 

MATCH2 VARCHAR2(10) 

) 

INSERT INTO MATCHTABLE VALUES('A1','A2'); 










SELECT * FROM MATCHTABLE; 

MATCH1 MATCH2 

----- ----- 

A1 A2 

A2 A1 

A2 A3 

A3 A2 

A4 A5 

A5 A4 

A4 A6 

A6 A4 

A7 A8 

A8 A7 

In this table, we will always have 2 rows for corresponding any 2 matchids. I need the output 

as below grouped matchids in groups. The problem I am facing is there is no starting point from 

where to start the hierarchy tree. And if I start matches for all distinct matchids, the groups 

tend to be the subset of some other superset and I need the supersets only. 

GROUPID MATCHID 

--------------------- 

1 A1 

1 A2 

1 A3 

2 A4 

2 A5 

2 A6 

3 A7 

3 A8 

Thanks.


you need to explain this better. if you have nothing to "start with", no way to describe what to "start with" - how can I figure out 

"what to start with" 

give me some logic here - FORGET CONNECT BY - it might not even be the proper approach. Describe the inputs and the 

DESIRED OUTPUTS using english, in the form of a program specification - pretend you are the end user (not a 

programmer) and you want a programmer to develop code for you - what would you as the end user say to the programmer 

to describe what you want. 

Connect By Clause April 6, 2009 - 2am Central time zone Bookmark | Bottom | Top 

Reviewer: Chandra from India 

Sorry... 

From the user perspective, table MAtchTable contains cols Match1 and Match2 which are the match 

ids. Any row eg A1, A2 corresponds that A1 and A2 ids have got a match among them. So, I need to 

group the matchids. 

We start picking the matches from column Match1- A1, A1 = A2 and then for A2, A2 = (A3 and A1) and 

then for A3, A3 = A2. So by avoiding the circular loop, A1, A2 and A3 falls in a single group. Note 

- Circular loops are to be avoided. 

If we start from A2, A2=(A1 and A3) and then for A1, A1=A2 and for A3, A3=A2. So they are again 

grouped as A1, A2 and A3. 

Similarly, A4,A5 and A6 share common matches, so are grouped together. 

And so with A7 and A8. 

May 19, 2009 - 10am Central time zone Bookmark | Bottom | Top 

Reviewer: Balaji from India 

Hi tom, 

Can you explain how the below mentioned query execute, i am confused with 

second connect by prior condition in query2 and by adding this the query get executed fast. 

what is the difference between query1 and query2 in execution 

query1:- 

======== 

SELECT lm_parentlimit_lm, lm_credprogram_cp_k, lm_limit_k, LEVEL 

FROM ca_lmcreditprogramlimits 

WHERE lm_credprogram_cp_k = 1088 AND lm_parentlimit_lm IS NULL 

START WITH lm_limit_k = 22 

CONNECT BY PRIOR lm_parentlimit_lm = lm_limit_k 

query2:- 

======== 

SELECT lm_parentlimit_lm, lm_credprogram_cp_k, lm_limit_k, LEVEL 

FROM ca_lmcreditprogramlimits 

WHERE lm_credprogram_cp_k = 1088 AND lm_parentlimit_lm IS NULL 

START WITH lm_limit_k = 22 

CONNECT BY PRIOR lm_parentlimit_lm = lm_limit_k 

AND PRIOR lm_credprogram_cp_k = lm_credprogram_cp_k 

Followup May 23, 2009 - 12pm Central time zone: 

they are as different as night and day 

you would actually need to understand the constructs you are using here.... do you understand the use of connect by, when 

and how the where clause is applied? that the two queries you present are ENTIRELY AND UTTERLY different - they do not 

return the same answers? 

They each are valid. 

But they each answer entirely different queries, different questions. 

One tells you why the sky is blue, the other why grass is green (as way of analogy)

A connect by with a where clause is similar to : 

a) build the ENTIRE HIERARCHY 

b) then apply the where clause 

A connect by - by itself is 

a) build the ENTIRE HIERARCHY 

now, your query two, it builds (stops faster) a MUCH MUCH SMALLER hierarchy. 

The query one, builds a big hierarchy and then filters it. 

you cannot compare the above two queries, they are as different as 

select * from emp; 

select * from dept; 

are. 

Strange behavior December 22, 2010 - 9am Central time zone Bookmark | Bottom | Top 

Reviewer: Mihail Bratu from Romania 

Hi Tom, 

I discovered some strange behavior around the connect by clause. I'll reproduce this issue into one 

small test case. 

drop table emp 

/ 

drop table enames 

/ 

create table emp as select * from scott.emp 

/ 

create table enames(ename varchar2(10)) 

/ 

insert into enames 

select column_value from table(sys.odcivarchar2List('KING','JONES','CLARK')) 

/ 

commit 

/ 

Let's see the tree starting from mgr is null and including the employee names from the enames 

table: 

column ename_s format a20 

select rpad(' ' ,2*(level-1), ' ')|| e.ename ename_s, prior e.ename ename_prior 

from emp e 


and e.ename in (select ename from enames) 


/ 

ENAME_S ENAME_PRIO 

-------------------- ---------- 

KING 

JONES KING 

CLARK KING 

Now we'll evolve the tree applying the condition on connect by with one level delay using the prior 

operator. 

We expect to obtain the above tree with one level extended. Instead of that the whole tree is 

returned. 


from emp e 


and prior e.ename in (select ename from enames)


/ 


-------------------- ---------- 

KING 

JONES KING 

SCOTT JONES 

ADAMS SCOTT 

FORD JONES 

SMITH FORD 

BLAKE KING 

ALLEN BLAKE 

WARD BLAKE 

MARTIN BLAKE 

TURNER BLAKE 

JAMES BLAKE 

CLARK KING 

MILLER CLARK 

Now we'll use the EXISTS function for the same target: 


from emp e 


and prior case when exists( 

select null from enames where ename = e.ename 

) then 1 end = 1 


/ 


-------------------- ---------- 

KING 

JONES KING 

SCOTT JONES 

FORD JONES 

BLAKE KING 

CLARK KING 

MILLER CLARK 

The result matches the expectation! 

The question is, is this a bug, or I'm missing something? I appreciate your comments. 

Thank you 

Followup December 22, 2010 - 2pm Central time zone: 

In 10gR2 - I reproduce your findings (10.2.0.4) 

in 11gR2 - I do NOT reproduce your finds, it does the right thing (11.2.0.2) 

please contact support for this one. 

Thank you (11.1.0.6.0) December 23, 2010 - 2am Central time zone Bookmark | Bottom | Top 

Reviewer: Mihail Bratu 

connect by prior with nulls?? January 21, 2011 - 5am Central time zone Bookmark | Bottom | Top 


Hi Tom, 

In our 2 main DBs we have the following code 

select sysid,party_link_sysid 

from parties 

CONNECT BY PRIOR SYSID = PARTY_LINK_SYSID 

start with sysid=:a1; 

both with the same indexes. However the data is different 

DB1 

-------

select count(*) from parties; 

COUNT(*) 

---------- 

7377820 

select count(party_link_sysid) from parties; 

COUNT(PARTY_LINK_SYSID) 

----------------------- 

1533750 

DB2 

--------- 

select count(*) from parties; 

COUNT(*) 

---------- 

1259267 

select count(party_link_sysid) from parties; 

COUNT(PARTY_LINK_SYSID) 

----------------------- 

34 

results from autot for DB1 


-------------------------------------------------------------------------------------------------- 



| 2 | TABLE ACCESS BY INDEX ROWID | PARTIES | 1 | 24 | 1 (0)| 00:00:01 | 

|* 3 | INDEX UNIQUE SCAN | PARTY_PK | 1 | | 1 (0)| 00:00:01 | 

| 4 | NESTED LOOPS | | | | | | 


| 6 | TABLE ACCESS BY INDEX ROWID| PARTIES | 8 | 72 | 1 (0)| 00:00:01 | 

|* 7 | INDEX RANGE SCAN | PARTY_PARTY_FK_I | 2 | | 1 (0)| 00:00:01 | 

results from autot for DB2 

| Id | Operation | Name | Rows | Bytes | Cost | 

---------------------------------------------------------------------------------- 

| 0 | SELECT STATEMENT | | 63036 | 492K| 1 | 

| 1 | CONNECT BY WITH FILTERING | | | | | 

| 2 | TABLE ACCESS BY INDEX ROWID | PARTIES | 1 | 22 | 1 | 

| 3 | INDEX UNIQUE SCAN | PARTY_PK | 1 | | 1 | 

| 4 | NESTED LOOPS | | | | | 

| 5 | CONNECT BY PUMP | | | | | 

| 6 | TABLE ACCESS BY INDEX ROWID| PARTIES | 63036 | 492K| 1 | 

| 7 | INDEX RANGE SCAN | PARTY_PARTY_FK_I | 1 | | 1 | 

---------------------------------------------------------------------------------- 

So it seems like the DB with 33 values and the rest of the table is null for the party_link_sysid 

column, gets half the table scanned. 

Is there a way to get it to scan a smaller amount - like in DB1? 



those are explain plans, they are not showing you what actually happened when you ran the query, they are showing a 

guess of what might happen. 

there are no scans here, only indexed reads. 

It looks like your statistics are not up to date somewhere and when they get up to date - you'll see different estimates.

It looks like your statistics are not up to date somewhere and when they get up to date - you'll see different estimates. 

Use sql_trace+tkprof to see what ACTUALLY happens row wise. 

connect by prior January 24, 2011 - 8am Central time zone Bookmark | Bottom | Top 


Thanks tom - should we ignore the Rows column in autot ? 

What is it for? 

I have analayzed all indexes and tables involved - but the explain plans come out the same: 

DB1 

------------------ 

SQL> select table_name,num_rows,last_analyzed from dba_tables 

2 where table_name ='PARTIES'; 

TABLE_NAME NUM_ROWS LAST_ANAL 

------------------------------ ---------- --------- 

PARTIES 7378389 24-JAN-11 

SQL> select index_name, num_rows,last_analyzed from dba_indexes 


INDEX_NAME NUM_ROWS LAST_ANAL 

------------------------------ ---------- --------- 

PARTY_PK 7378389 24-JAN-11 

PARTY_OPT1 7378389 24-JAN-11 

PARTY_PARTY_FK_I 1533762 24-JAN-11 

PARTY_PARTY_FK_I2 245157 24-JAN-11 

PARTY_OPT2 7372722 24-JAN-11 

-------------------------------------------------------------------------------------- 

| Id | Operation | Name | Rows | Bytes | Cost (%CPU) 

-------------------------------------------------------------------------------------- 

| 0 | SELECT STATEMENT | | 8 | 72 | 1 (0) 

|* 1 | CONNECT BY WITH FILTERING | | | | 

| 2 | TABLE ACCESS BY INDEX ROWID | PARTIES | 1 | 24 | 1 (0) 

|* 3 | INDEX UNIQUE SCAN | PARTY_PK | 1 | | 1 (0) 

| 4 | NESTED LOOPS | | | | 

| 5 | CONNECT BY PUMP | | | | 

| 6 | TABLE ACCESS BY INDEX ROWID| PARTIES | 8 | 72 | 1 (0) 

|* 7 | INDEX RANGE SCAN | PARTY_PARTY_FK_I | 2 | | 1 (0) 

-------------------------------------------------------------------------------------- 

DB2 

------------------ 

SQL> select table_name,num_rows,last_analyzed from dba_tables 


TABLE_NAME NUM_ROWS LAST_ANAL 

------------------------------ ---------- --------- 

PARTIES 1208948 21-JAN-11 

SQL> select index_name, num_rows,last_analyzed from dba_indexes 


INDEX_NAME NUM_ROWS LAST_ANAL 

------------------------------ ---------- --------- 

PARTY_PK 1208948 21-JAN-11 

PARTY_OPT1 1208948 21-JAN-11 

PARTY_PARTY_FK_I 22 21-JAN-11 

PARTY_PARTY_FK_I2 6 21-JAN-11 

PARTY_OPT2 1203585 21-JAN-11 

| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| 

--------------------------------------------------------------------------------------- 

| 0 | SELECT STATEMENT | | 60447 | 472K| 1 (0)| 

|* 1 | CONNECT BY WITH FILTERING | | | | | 

| 2 | TABLE ACCESS BY INDEX ROWID | PARTIES | 1 | 22 | 1 (0)| 

|* 3 | INDEX UNIQUE SCAN | PARTY_PK | 1 | | 1 (0)| 

| 4 | NESTED LOOPS | | | | | 

| 5 | CONNECT BY PUMP | | | | |

| 6 | TABLE ACCESS BY INDEX ROWID| PARTIES | 60447 | 472K| 1 (0)| 

|* 7 | INDEX RANGE SCAN | PARTY_PARTY_FK_I | 1 | | 1 (0)| 

--------------------------------------------------------------------------------------- 

Does this difference not matter? Both queries timewise are the same - but as soon as this query is 

joined to other tables - the DB2 performs 30 times slower. 


... should we ignore the Rows column in auto ... 

NO, definitely not - that is the most important thing in an explain plan!!! 

You want to compare it to ACTUAL row counts from a sql_trace when you suspect an inefficient plan is being generated. If 

the ACTUALS differ widely from the GUESS (in the plan) - then that is the likely cause of the wrong plan being generated 

and we need to figure out "why" that is happening in order to correct it. 

Please - as stated above - sql_plus+tkprof if you want to see what they are ACTUALLY DOING. 

As you can see - the data sets are completely and utterly different. One of them seems to return A LOT more data. That 

might be a clue as to why they perform differently when used as a source to join to some other table. 

LEVEL, CONNECT BY February 3, 2011 - 3pm Central time zone Bookmark | Bottom | Top 

Reviewer: Ananth from Richmomn, VA USA 

Hi Tom, 

we have a table "folders" in content management system. 

Columns: 

folderid number 

levl number 

name varchar2 

pathids varchar2 

primary key is on folderid. 

Data: 

folderid levl name pathids 

0 0 Root :0: 

1 1 Test1 :0:1: 

2 1 Test2 :0:2: 

3 2 Test3 :0:1:3: 

4 3 Test4 :0:1:3:4: 

i want to have result 

folderid levl name path-name 

------------------------------- 

0 0 Root Root 

1 1 Test1 Root,Test1 

2 1 Test2 Root,Test2 

3 2 Test3 Root,Test1,Test3 

.. 

.. 

Also, when i tried to run a sample query i get different results 

SELECT FOLDERID,LEVL,LEVEL, 

ROW_NUMBER() OVER (PARTITION BY SNO ORDER BY SNO) RWN, 

PATHIDS 

FROM 

( 

SELECT FOLDERID, LEVL, NAME, PATHIDS 

FROM PATH 

WHERE FOLDERID IN (1) 

) 

CONNECT BY LEVEL

i get the output as 

FOLDERID LEVL LEVEL RWN 

---------- ---------- ---------- ---------- 

1 1 2 1 

1 1 1 2 

why did the level column showed in decreasing order. 

if i replace it with FOLDERID in (9) instead of (1) 

i get the results as 

FOLDERID LEVL LEVEL RWN 

---------- ---------- ---------- ---------- 

9 4 1 1 

9 4 2 2 

9 4 5 3 

9 4 4 4 

9 4 3 5 

am not able to figure it out. 

Can you please help me out in the above 2 scenario's 


no create table 

no inserts 

I did not even look at the question. 

Level, Connect By February 3, 2011 - 5pm Central time zone Bookmark | Bottom | Top 

Reviewer: Ananth from Richmomn, VA USA 

Hi Tom, 

Please find below the create table and insert scripts for the same. 

CREATE TABLE FOLDER 

( 

FOLDERID NUMBER NOT NULL PRIMARY KEY, 

LEVL NUMBER NOT NULL, 

NAME VARCHAR2(200) NOT NULL, 

PATHIDS VARCHAR2(200) NOT NULL 

); 

INSERT INTO FOLDER VALUES (0, 0, 'ROOT', ':0:'); 

INSERT INTO FOLDER VALUES (1, 1, 'TEST1', ':0:1:'); 

INSERT INTO FOLDER VALUES (2, 2, 'TEST2', ':0:1:2:'); 


INSERT INTO FOLDER VALUES (4, 3, 'TEST4', ':0:1:3:4:'); 



INSERT INTO FOLDER VALUES (7, 3, 'TEST7', ':0:1:6:7:'); 

INSERT INTO FOLDER VALUES (8, 4, 'TEST8', ':0:1:6:7:8:'); 

INSERT INTO FOLDER VALUES (9, 4, 'TEST9', ':0:1:6:7:9:'); 

I need a query which shows output as 

FOLDERID LEVL NAME PATH 

---------- ---------- ---------- ---------- 

0 0 ROOT ROOT 

1 1 TEST1 ROOT:TEST1 

2 2 TEST2 ROOT:TEST2 

.. 

.. 

.. 

also, for the below query, on why do the results (column LEVEL) are altered when i add analytic 

function. With out analytic function the level column shows as expected. (increasing order of 

LEVEL). 

SELECT 

FOLDERID, 

LEVL, 

LEVEL, 

ROW_NUMBER() OVER (PARTITION BY FOLDERID ORDER BY FOLDERID) RWNUM,

ROW_NUMBER() OVER (PARTITION BY FOLDERID ORDER BY FOLDERID) RWNUM, 

PATHIDS 

FROM 

( 

SELECT FOLDERID, LEVL, NAME, PATHIDS 

FROM FOLDER 

WHERE FOLDERID = 9 

) 

CONNECT BY LEVEL


that is an impossibly bad way to store a hierarchy - putting it back to together is impossible (well, not impossible but so 

inefficient as to make it be realistically "not something to do") 

We have to substr out the parent id (levl does *nothing* for us to put the data back together - nothing) and the connect with it. 

It isn't going to be "fast" on non-trivial data sets: 

ops$tkyte%ORA11GR2> select id, rpad('*',2*level,'*')||name nm, pathids, pid, 

2 sys_connect_by_path(name,',') scbp 

3 from ( 

4 select folderid id, name, pathids, 

5 substr( pathids, instr( pathids, ':', -1, 3 )+1, 

instr(pathids,':',-1,2)-instr(pathids,':',-1,3)-1 ) pid 

6 from folder 

7 ) 


9 connect by prior id = pid 

10 / 

ID NM PATHIDS PID SCBP 

---------- -------------------- -------------------- -------------------- 

------------------------------ 

0 **ROOT :0: ,ROOT 

1 ****TEST1 :0:1: 0 ,ROOT,TEST1 

2 ******TEST2 :0:1:2: 1 ,ROOT,TEST1,TEST2 


4 ********TEST4 :0:1:3:4: 3 ,ROOT,TEST1,TEST3,TEST4 



7 ********TEST7 :0:1:6:7: 6 ,ROOT,TEST1,TEST6,TEST7 

8 **********TEST8 :0:1:6:7:8: 7 

,ROOT,TEST1,TEST6,TEST7,TEST8 

9 **********TEST9 :0:1:6:7:9: 7 

,ROOT,TEST1,TEST6,TEST7,TEST9 


you should forget about storing levl and pathids, just store PID - the parent id, the others are all easily derived from the 

connect by hierarchy. 

Connect BY Prior April 5, 2011 - 2pm Central time zone Bookmark | Bottom | Top 

Reviewer: Ananth from Richmod, VA USA 

Hi Tom, 

I have a query regarding the Hierarchial queries. 

Is there any way to differentiate between each hierarchy. i.e i want to logically differentiate 

each hierarchy ie. (leaf to root), just as how we logically differentiate regions by using group by 

region. 

can you give me some ideas or suggestions on how to achieve the same. 

Regards 

Ananth 


have you looked at sys_connect_by_path - maybe that'll help you. 

I'm not really sure what you mean otherwise, I don't know how to compare group by with a connect by? Do you have an 

example? 

April 12, 2011 - 5pm Central time zone Bookmark | Bottom | Top 

Reviewer: Ananth from Richmond, VA USA 

Hi Tom, 

I could think of CONNECT_BY_ROOT as one option which can differentiate a hierarchy within.

for eg: 

in traditional employee table if you go from Bottom - Top approach, you have in a hierarchy Employee-his managermanager's 

manager and so on till employee with manager_id null. 

So if i say that Leaf to Root as one hierarchy. is it possible to do some aggregate kind of things on each hierarchy. 

from Traditional employee table design, i see employee_id being unique, i can do the aggregate function on hierarchies 

using connect_by_root employee_id. 

In some scenarions where we couldnt differentiate a hierarchy.. how can we do on those..? 

Regards 

Ananth 


give me an example please - use scott.emp. I don't know what you want to "aggregate" - you know how to get the root - so 

aggregating is easy, you have something to group by. But I suspect you mean something OTHER than aggregation don't 

you. 

give an EXAMPLE of what you are looking for - be specific, explain everything. 

SYS_CONNECT_BY_PATH April 14, 2011 - 12am Central time zone Bookmark | Bottom | Top 


Hi Tom, 

Lets say i have the path from Root is stored in a table. 

For Ex: in Traditional EMPLOYEES table, i have only columns as 

EMPLOYEE_ID, EMPLOYEE_PATH (assuming i dont have hierarchy information). 

 

EMPLOYEE_ID EPATH 

----------- -------------- 

101 :101 

108 :101:108 

109 :101:108:109 

111 :101:108:111 

112 :101:108:112 

113 :101:108:113 

110 :101:108:110 

204 :101:204 

205 :101:205 

206 :101:205:206 

203 :101:203 

200 :101:200 

102 :102 

103 :102:103 

104 :102:103:104 

107 :102:103:107 

106 :102:103:106 

105 :102:103:105 

114 :114 

115 :114:115 

Question1: 

----------- 

how do i get the below output 

EMPLOYEE_ID PATH SERIAL 

101 101 1 

108 101 1 

108 108 2 

109 101 1 

109 108 2 

109 109 3 

.. 

.. 

.. 

 

Followup April 14, 2011 - 9am Central time zone:

this can be done, but what you have to do first is provide 

create table 

inserts 

then I can supply select 

actually, you can supply the select, here is the technique: 


4 

Scripts April 14, 2011 - 10am Central time zone Bookmark | Bottom | Top 


Hi Tom, 

PFB the scripts 

create table emp 

( 

id number, 

path varchar2(100) 

); 

insert into emp values (101,':101:'); 

insert into emp values (108,':101:108:'); 

insert into emp values (109,':101:108:109'); 

insert into emp values (111,':101:108:111:'); 


















tell you what Ananth, give it what we call the "old college try" using the link I provided first. You should be able to - they are 

virtually identical in nature. 

And if you cannot - come back and show your work and we'll take it from there. 

April 14, 2011 - 10am Central time zone Bookmark | Bottom | Top 


Sure Tom, Am actually looking into it. 

wil surely let you know my approach if i couldnt be able to make it. 

ask why conect by didn't work at oracle 11g (11.2.0.1.0) July 22, 2011 - 5am Central time zone 


Reviewer: Liliek from INdonesia 

Hi Tom, 

i want to ask you : 

i have been migrate from oracle 10.2.0.1.0 ver to 11.2.0.1.0 and this start with.. connect by.. 

clause not work. could you help me to solve it, because i have a report run by this query for one 

and half hours!!

and half hours!! 

thx tom , i really appreciate for your help 

Rgds 

Liliek 


umm, connect by and start with most certainly DO WORK with 11g 

you'd need to be a bit more precise in describing your problem 

PLESE SOLVE MY DOUBT March 30, 2012 - 4am Central time zone Bookmark | Bottom | Top 

Reviewer: HANMATH PRADEEP from INDIA 

Hello every1 ……………….. I have a small doubt regarding retrieving records from the base table….. So 

please clarify my doubt …………………………….. 

The respective base table name is “PRADEEP”. 

SQL>SELECT *FROM PRADEEP; 

SNO NAME ADDR 

------------------------------------ 

1 HANMATH PRADEEP HYD 

2 HARI PRASAD CHENNAI 

3 HARI SHANKER BANGLORE 

4 HARINATH NAIDU PAKISTAN 

5 HARI PRASAD GUNTUR 

6 SURESH PAIDY VIZAG 

EXPECTING OUTPUT SHOULD BE IN THE FOLLOWING MANNER: 

SNO NAME ADDR 

1 HANMATH PRADEEP HYD 

2 HARI PRASAD CHENNAI 

5 HARI PRASAD GUNTUR 

MY REQUIREMENT IS AS FOLLOWS: 

(1)……… HERE I WOULD LIKE TO RETRIEVE THE RECORDS BASED ON THE NAME COLUMN ONLY. 

(2) I KNOW THIS QUERY I.E., (SELECT *FROM PRADEEP WHERE NAME LIKE ‘H_______P%’ OR NAME LIKE 

‘H____P%’;) ……………… I DON’T LIKE TO USE SUCH LIKE STATEMENTS HERE? SO PLZ AVOID IT… 

(3)……….. THE IMPORTANT CONDITION IS THAT …. THE RECORDS IN WHICH THE STARTING LETTER IS “”H”” (IN 

MIDDLE NAME) & THE STARTING LETTER IS “”P”” (IN LAST NAME) SHOULD ONLY RETRIEVED FROM THE BASE 

TABLE. 


every1? plz? 

what is this? elementary school? 

sigh, i did read it - nothing to be read here. "I don't like to use such like statements". to which I say "sorry, but what would you 

like to use then" 

or where you looking for "name like 'H% P%'" - starts with an H and contains a P preceded by a space. but that presumes, 

assumes that everyone just has a single word first name and single word last name and nothing else (a rather naive 

assumption) but you give us no details. 

(did you know your capslock is apparently stuck?) 


Reviewer: sym from US

Is there any way to fold a hierarchy into a collection? 

Suppose I have 

create type x as object(val varchar2(100)) not final; 

create type x_list as table of x; 

create type y under x(children x_list); 

create table z ( 

val varchar2(100), 

id int, 

parent_id int); 

insert into z values('a',1, null); 

insert into z values('b',2, 1); 

insert into z values('c',3, 2); 

Is there a way to extract this into x with the following structure 

x('a',x_list(x('b',x_list(x('c',NULL))))); 



Reviewer: sym from US 

The previous question, it is a y object and not x that needs to be constructed. Sorry for the 

mistake. 

Write a Review 

All information and materials provided here are provided "as-is"; Oracle disclaims all express and implied warranties, including, the 

implied warranties of merchantability or fitness for a particular use. Oracle shall not be liable for any damages, including, direct, indirect, 

incidental, special or consequential damages for loss of profits, revenue, data or data use, incurred by you or any third party in connection 

with the use of this information or these materials. 

About Oracle | Legal Notices and Terms of Use | Privacy Statement

Thanks for the question regarding "connect by ", versi

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?