optimizing insert query - sql

Question

I am trying to insert comma separated values in a global table. When data is large it's taking long time to process the data. I need to optimize my insert query, is there any other ways to achieve below insert statement for better optimization? Please check code below for more info. Appreciated for any help.
//my proc
emp_id in CLOB;
//insert statement
insert into Global_Emp_Tbl
with inputs(str) as(
select to_clob(emp_id)
from dual
),
temp_table(s, n, empid, st_pos, end_pos) as (
select ',' || str || ',', -1, null, null, 1
from inputs
union all
selct s, n+1, substr(s, st_pos, end_pos - st_pos),
end_pos + 1, instr(s, ',', 1, n+3)
from temp_table
where end_pos != 0
)
select empid from temp_table where empid is not null;
commit;
//using insert table in where clause
exists( select 1 from Global_Emp_Tbl gt where e.id =gt.emp_id ) //joining with main table

Answer 1

You can use simple REGEXP_SUBSTR to achieve the same
insert into Global_Emp_Tbl
SELECT Regexp_substr(empid, '[^,]+', 1, LEVEL) AS empid
FROM (SELECT To_clob(emp_id) empid
FROM dual)
CONNECT BY LEVEL <= Regexp_count(emp_id, '[^,]+');
commit;
Also, one more suggestion try changing the below in your function
select empid from temp_table where n > 0;

Answer 2

Well I can't reporduce your problem in Oracle Database 11g Express Edition Release 11.2.0.2.0 - 64bit Production
This procedure provides the sample data, a CLOB csv list of the first N interger
create or replace function get_list(N NUMBER) return CLOB
as
v_lst CLOB;
i PLS_INTEGER;
BEGIN
v_lst :='1';
for i in 2 .. n loop
v_lst :=v_lst||','||to_char(i);
end loop;
return(v_lst);
END;
/
Note that the 5K parameter give approximately 20K long list.
select length(get_list(5000)) from dual;
23892
The parsing of this list in a global temporary table is done in seconds not in minutes. Here an example using your SELECT
SQL> set timi on;
SQL> create global temporary table csv_tbl
2 ON COMMIT PRESERVE ROWS
3 as
4 with inputs(str) as(
5 select get_list(5000)
6 from dual
7 ),
8 temp_table(s, n, empid, st_pos, end_pos) as (
9 select ',' || str || ',', -1, null, null, 1
10 from inputs
11 union all
12 select s, n+1, substr(s, st_pos, end_pos - st_pos),
13 end_pos + 1, instr(s, ',', 1, n+3)
14 from temp_table
15 where end_pos != 0
16 )
17 select empid from temp_table where empid is not null
18 ;
Table created.
Elapsed: 00:00:01.35
So the most probably explanation is, that the most elapsed time is spend in the query with the EXISTS clause
exists( select 1 from Global_Emp_Tbl gt where e.id =gt.emp_id )
Two problems come to my mind
1) The datatypes of the EMP_ID in the temporary table and in the EMP table differs, e.g. in temporary table it is VARCHAR2, in the EMP table NUMBER - this will prohibit the use of the index on the EMP table.
2) The missing object statistics on the global temporary table leads the CBO to an wrong execution plan and you use the index (in a large nested loops) where you should use a full table scan.