I'm having a problem while trying to apply a filter in an Oracle (12.2.0.1.0) query.
When I query for the values of a column in a table using order by that same column, I get
75A0000000
7597953181
7597953182
But then, when I try to filter the rows with the column between the first and the last value I'm not getting results at all.
This works perfectly fine if the column has only numeric values, but it gets it all wrong when there is some alphabetic value in the middle.
It seems like Oracle is following the criteria A<0 when ordering a result but A>0 when filtering that result.
This can be reproduced using this query:
select * from (
WITH DATA AS
(SELECT '1, A, 2' str FROM dual)
SELECT trim(regexp_substr(str, '[^,]+', 1, LEVEL)) str
FROM DATA
CONNECT BY instr(str, ',', 1, LEVEL - 1) > 0)
where str>='1' and str<='A'
order by str
The result of this query is
A
1
2
that should mean A is less than 1, but the filter applied to make it return some values is just the opposite : str values greater or equal than 1 and less or equal than A.
Anybody knows why? Thanks in advance.
To me, it looks as if it is about binary or linguistic sort.
Sample data:
SQL> select * from test;
C
-
1
A
2
Sort by col, as is:
SQL> select col, ascii(col)
2 from test
3 order by col;
C ASCII(COL)
- ----------
A 65
1 49
2 50
Or, sort by col's ASCII code:
SQL> select col, ascii(col)
2 from test
3 order by ascii(col);
C ASCII(COL)
- ----------
1 49
2 50
A 65
Explicitly stating binary or linguistic sort:
SQL> select col
2 from test
3 order by nlssort(col, 'nls_sort=binary'); --> binary
C
-
1
2
A
SQL> select col
2 from test
3 order by nlssort(col, 'nls_sort=croatian'); --> linguistic
C
-
A
1
2
SQL>
Or, alter session and set nls_sort you want:
SQL> alter session set nls_sort='binary';
Session altered.
SQL> select * From test order by col;
C
-
1
2
A
SQL>
If you get unexpected result, check what's nls_sort set in your session:
SQL> select * from nls_session_parameters where parameter = 'NLS_SORT';
PARAMETER VALUE
------------------------------ ----------------------------------------
NLS_SORT BINARY
SQL>
Using your query:
SQL> select * from (
2 WITH DATA AS
3 (SELECT '1, A, 2' str FROM dual)
4 SELECT trim(regexp_substr(str, '[^,]+', 1, LEVEL)) str
5 FROM DATA
6 CONNECT BY instr(str, ',', 1, LEVEL - 1) > 0)
7 where str>='1' and str<='A'
8 order by nlssort(str, 'nls_sort=binary'); --> this
STR
-------
1
2
A
SQL>
CREATE OR REPLACE PROCEDURE UPDATE_CRDT_JV IS
BEGIN
UPDATE GL_DISTRIBUTION
SET GL_DATE = (SELECT ADJ_DATE FROM ADJUSTMENTS WHERE
ADJ_NUMBER = TO_NUMBER(TR_NUMBER))
WHERE TR_TYPE = 'ADJST';
UPDATE GL_DISTRIBUTION
SET GL_DATE = (SELECT PARTY_ADJ_DATE FROM PARTY_ADJUSTMENT
WHERE PARTY_ADJ_NUMBER = TO_NUMBER(TR_NUMBER))
WHERE TR_TYPE = 'PRTAJ';
UPDATE GL_DISTRIBUTION
SET GL_DATE = (SELECT VEN_PAY_VOU_DATE FROM PAYMENTS_TO_VENDORS WHERE
VEN_PAY_VOU_NUMBER = TO_NUMBER(TR_NUMBER))
WHERE TR_TYPE = 'CRPAY';
UPDATE GL_DISTRIBUTION
SET GL_DATE = (SELECT CHEQUE_DATE FROM SYS_PAYMENTS_HEADER WHERE
REF_NUMBER = TO_NUMBER(TR_NUMBER))
WHERE TR_TYPE = 'SYSPY';
UPDATE GL_DISTRIBUTION
SET GL_DATE = (SELECT POSTED_DATE FROM PURCHASE_INVOICE_HEADER WHERE
POSTED_DATE IS NOT NULL AND PIV_NUMBER = TO_NUMBER(TR_NUMBER))
WHERE TR_TYPE = 'CRINV';
UPDATE GL_dISTRIBUTION
SET GL_dATE = (SELECT DOC_dATE FROM REVERSE_HISTORY
WHERE TR_NUMBER = TO_NUMBER(GL_DISTRIBUTION.TR_NUMBER)
AND DOC_dATE IS NOT NULL AND TR_TYPE IN ('SYSPY','CRPAY'))
WHERE TR_TYPE IN ('RSYSPY','RCRPAY');
commit;
UPDATE_INV_DET;
END;
is taking more than 15 minutes to update the database.
now i am updating this by using the following query in SQL PLUS:
EXECUTE UPDATE_CRDT_JV;
pls help me if any body knows the solution for this problem
I agree with the already given advice to start figuring out where your code is spending time. However, your case is quite common and I think I recognize this situation: you have coded your update statements in such a way that the other tables are accessed for every row of GL_DISTRIBUTION of that type.
The solution is to rewrite your update statements and I see two possibilities to do that efficiently:
1) Update a select statement (UPDATE (SELECT ...) SET ... WHERE ...). This requires some unique key constraints to be in place or using the BYPASS_UJVC hint.
2) Use a MERGE statement.
Below you see an example of how to rewrite your code using a single merge statement. I'm expecting big performance gains because the access of the other tables is now done once using a single outer join for each of the tables, instead of for every row in the GL_DISTRIBUTION table.
The example. Test data:
SQL> create table gl_distribution (tr_number, tr_type, gl_date)
2 as
3 select '1', 'ADJST', date '2011-01-01' from dual union all
4 select '2', 'ADJST', null from dual union all
5 select '3', 'PRTAJ', date '2011-01-01' from dual union all
6 select '4', 'SYSPY', date '2011-01-01' from dual union all
7 select '5', 'RCRPAY', date '2011-01-01' from dual
8 /
Table created.
SQL> create table adjustments (adj_number, adj_date)
2 as
3 select 1, sysdate from dual union all
4 select 2, sysdate from dual
5 /
Table created.
SQL> create table party_adjustment (party_adj_number, party_adj_date)
2 as
3 select 3, sysdate from dual union all
4 select 33, sysdate from dual
5 /
Table created.
SQL> create table payments_to_vendors (ven_pay_vou_number, ven_pay_vou_date)
2 as
3 select 34, sysdate from dual
4 /
Table created.
SQL> create table sys_payments_header (ref_number,cheque_date)
2 as
3 select 4, sysdate from dual
4 /
Table created.
SQL> create table purchase_invoice_header (piv_number,posted_date)
2 as
3 select 35, sysdate from dual
4 /
Table created.
SQL> create table reverse_history (tr_number,doc_date,tr_type)
2 as
3 select 5, sysdate, 'CRPAY' from dual
4 /
Table created.
SQL>
Your procedure (for comparison):
SQL> CREATE OR REPLACE PROCEDURE UPDATE_CRDT_JV
2 IS
3 BEGIN
4 UPDATE GL_DISTRIBUTION
5 SET GL_DATE = (SELECT ADJ_DATE FROM ADJUSTMENTS WHERE ADJ_NUMBER = TO_NUMBER(TR_NUMBER))
6 WHERE TR_TYPE = 'ADJST'
7 ;
8 UPDATE GL_DISTRIBUTION
9 SET GL_DATE = (SELECT PARTY_ADJ_DATE FROM PARTY_ADJUSTMENT
10 WHERE PARTY_ADJ_NUMBER = TO_NUMBER(TR_NUMBER))
11 WHERE TR_TYPE = 'PRTAJ'
12 ;
13 UPDATE GL_DISTRIBUTION
14 SET GL_DATE = (SELECT VEN_PAY_VOU_DATE FROM PAYMENTS_TO_VENDORS
15 WHERE VEN_PAY_VOU_NUMBER = TO_NUMBER(TR_NUMBER))
16 WHERE TR_TYPE = 'CRPAY'
17 ;
18 UPDATE GL_DISTRIBUTION
19 SET GL_DATE = (SELECT CHEQUE_DATE FROM SYS_PAYMENTS_HEADER WHERE
20 REF_NUMBER = TO_NUMBER(TR_NUMBER))
21 WHERE TR_TYPE = 'SYSPY'
22 ;
23 UPDATE GL_DISTRIBUTION
24 SET GL_DATE = (SELECT POSTED_DATE FROM PURCHASE_INVOICE_HEADER WHERE
25 POSTED_DATE IS NOT NULL AND PIV_NUMBER = TO_NUMBER(TR_NUMBER))
26 WHERE TR_TYPE = 'CRINV'
27 ;
28 UPDATE GL_dISTRIBUTION
29 SET GL_dATE = (SELECT DOC_dATE FROM REVERSE_HISTORY
30 WHERE TR_NUMBER = TO_NUMBER(GL_DISTRIBUTION.TR_NUMBER)
31 AND DOC_dATE IS NOT NULL AND TR_TYPE IN ('SYSPY','CRPAY'))
32 WHERE TR_TYPE IN ('RSYSPY','RCRPAY')
33 ;
34 --commit;
35 --UPDATE_INV_DET;
36 END;
37 /
Procedure created.
SQL>
My suggestion:
SQL> create procedure new_update_crdt_jv
2 as
3 begin
4 merge into gl_distribution d
5 using ( select to_number(d.tr_number) tr_number
6 , coalesce
7 ( a.adj_date
8 , pa.party_adj_date
9 , pv.ven_pay_vou_date
10 , sph.cheque_date
11 , pih.posted_date
12 , rh.doc_date
13 ) new_date
14 from gl_distribution d
15 left outer join adjustments a
16 on to_number(d.tr_number) = a.adj_number
17 and d.tr_type = 'ADJST'
18 left outer join party_adjustment pa
19 on to_number(d.tr_number) = pa.party_adj_number
20 and d.tr_type = 'PRTAJ'
21 left outer join payments_to_vendors pv
22 on to_number(d.tr_number) = pv.ven_pay_vou_number
23 and d.tr_type = 'CRPAY'
24 left outer join sys_payments_header sph
25 on to_number(d.tr_number) = sph.ref_number
26 and d.tr_type = 'SYSPY'
27 left outer join purchase_invoice_header pih
28 on to_number(d.tr_number) = pih.piv_number
29 and d.tr_type = 'CRINV'
30 left outer join reverse_history rh
31 on to_number(d.tr_number) = rh.tr_number
32 and rh.tr_type in ('SYSPY','CRPAY')
33 and d.tr_type in ('RSYSPY','RCRPAY')
34 ) n
35 on ( d.tr_number = n.tr_number)
36 when matched then
37 update set d.gl_date = n.new_date
38 ;
39 end new_update_crdt_jv;
40 /
Procedure created.
SQL>
Let's run your procedure:
SQL> select * from gl_distribution
2 /
T TR_TYP GL_DATE
- ------ -------------------
1 ADJST 01-01-2011 00:00:00
2 ADJST
3 PRTAJ 01-01-2011 00:00:00
4 SYSPY 01-01-2011 00:00:00
5 RCRPAY 01-01-2011 00:00:00
5 rows selected.
SQL> exec update_crdt_jv
PL/SQL procedure successfully completed.
SQL> select * from gl_distribution
2 /
T TR_TYP GL_DATE
- ------ -------------------
1 ADJST 31-03-2011 14:41:19
2 ADJST 31-03-2011 14:41:19
3 PRTAJ 31-03-2011 14:41:19
4 SYSPY 31-03-2011 14:41:19
5 RCRPAY 31-03-2011 14:41:19
5 rows selected.
SQL> rollback
2 /
Rollback complete.
SQL>
My procedure returns the same results:
SQL> exec new_update_crdt_jv
PL/SQL procedure successfully completed.
SQL> select * from gl_distribution
2 /
T TR_TYP GL_DATE
- ------ -------------------
1 ADJST 31-03-2011 14:41:19
2 ADJST 31-03-2011 14:41:19
3 PRTAJ 31-03-2011 14:41:19
4 SYSPY 31-03-2011 14:41:19
5 RCRPAY 31-03-2011 14:41:19
5 rows selected.
Hope this helps.
Regards,
Rob.
The solution for this problem is:
Figure out where your code is
spending its time (i.e. profile it)
Figure out how to speed up the
slowest part
Repeat until performance is acceptable
If you prefer guesswork, then you might want to try any of the following:
Combine multiple UPDATEs into a single UPDATE statement, e.g. using a CASE condition as shown by #Aklopper.
Use MERGE instead of UPDATE to avoid correlated subqueries. Might be better, might not.
Look into the UPDATE_INV_DET procedure which is called at the end of the procedure shown.
I come from a SQL Server environment, won't a CASE UPDATE statement like this also help you(example of usage of the UPDATE CASE statement in SQL(don't know if Oracle has an equivalent methods) :
UPDATE titles
SET GL_DATE=
CASE
WHEN TR_TYPE = 'ADJST' THEN (SELECT ADJ_DATE FROM ADJUSTMENTS WHERE ADJ_NUMBER = TO_NUMBER(TR_NUMBER) END
WHEN TR_TYPE = 'PRTAJ'' THEN (SELECT PARTY_ADJ_DATE FROM PARTY_ADJUSTMENT
WHERE PARTY_ADJ_NUMBER = TO_NUMBER(TR_NUMBER))
END
ELSE price
END
I've got some help turning my table of the sort:
Col
23
25
15
53
...
into something like 23,25,15,53...
The query that does it is
SELECT max(ltrim(sys_connect_by_path(flow_run_id, ','), ','))
FROM
(select flow_run_id, rownum rn
from table
where CREATED_DATE < sysdate - 32
and flow_id = 3
order by 1 desc)
START WITH rn = 1
CONNECT BY PRIOR rn = rn - 1
(this beaulty was given by Michael in here)
My current problem is that the result is too long (ORA-01489 over the 4k chars from varchar2). I'm still learning about these sys_connected_by_path so I'd need some help sorting this. How could I make this query return me multiple rows instead of one super long line? i.e.:
Instead of
419,1,2,3,411,418,4,415,887,413,414,201,888,890,401,417,610,412,416,5,6,922,1080,1422,1423,1411,1412,1413,1414,1415,1416,1417,1418,1419,1964,2217,1636,2037,1988,1970,2038,1989,2000,2040,1993,2043,1994,2001,2044,1658,1995,2045,2224,1996,2019,1678,1997,2022,2201,1680,2219,2024,2207,1677,2209,2220,1959,2211,1961,2026,2212,1962,2028,2215,1675,1676,2035,2216,1986,1963,2017,1983,1935,2002,2018,1985,1936,2003,2020,2032,1937,2004,2021,2033,1938,1943,2023,2034,1939,1944,2025,2225,1941,1950,2027,2036,1942,1955,2029,2041,1945,1956,2030,2227,1946,1957,2031,2039,1947,2005,1974,2042,1948,2006,1976,2228,1949,2007,1978,1951,2009,1979,1929,1952,2012,1980,1931,1953,2013,1981,1933,1954,2015,2334,2350,2311,2239,2240,2241,2242,2245,2246,2249,2250,2336,2312,2008,2010,2011,2014,2251,2253,2016,2243,2244,2247,2351,2248,(...)
get
419,1,2,3,411,418,4,415,887,413,414,201,888,890,401,417,610,412,416,5,6,922,1080
1423,1411,1412,1413,1414,1415,1416,1417,1418,1419,1964,2217,1636,2037,1988,1970,2038
2000,2040,1993,2043,1994,2001,2044,1658,1995,2045,2224,1996,2019,1678,1997,2022,2201
(...)
Any tips?
Thanks!
f.
the following query will cut your big string in parts:
SQL> SELECT root_rn, MAX(concat)
2 FROM (SELECT connect_by_root(rn) root_rn,
3 ltrim(sys_connect_by_path(flow_run_id, ','), ',') concat
4 FROM (SELECT flow_run_id, rownum rn
5 FROM (SELECT round(dbms_random.VALUE(1, 10000))
6 AS flow_run_id
7 FROM dual
8 CONNECT BY ROWNUM <= 2000)
9 ORDER BY 1 DESC)
10 START WITH MOD(rn, 10) = 1
11 CONNECT BY PRIOR rn = rn - 1
12 AND MOD(rn, 10) != 1)
13 GROUP BY root_rn
14 ORDER BY root_rn;
ROOT_RN MAX(CONCAT)
---------- -------------------------------------------------------------------
1 654,6710,5297,5481,5085,2793,7646,9170,1051,2387
11 1882,8285,5430,4928,267,3779,3843,1151,3085,1446
21 4721,6087,6755,9904,805,2776,4633,2772,7785,5818
31 5189,5307,6481,2099,3832,9788,5970,8068,6605,3904
41 53,7013,1314,7717,9320,7069,907,5367,5013,7637
51 3903,2318,2611,7954,5751,5598,6148,6555,9724,984
[...]
You can replace "10" with a bigger number if you want more elements on each row.
Some little modifications to keep order
SELECT 10*frn+1 root,ltrim(sys_connect_by_path(flow_run_id,','),',') FROM
(SELECT flow_run_id,mod(rn,10) mrn,floor(rn/10) frn,count(*)over(partition by floor(rn/10))-1 crn FROM
(SELECT flow_run_id, row_number()over(order by flow_run_id)-1 rn FROM
(SELECT round(dbms_random.VALUE(1, 10000)) AS flow_run_id FROM dual CONNECT BY ROWNUM <= 2000
)
)
)
WHERE crn = mrn
START WITH mrn = 0
CONNECT BY PRIOR mrn = mrn-1 AND PRIOR frn = frn