Tune select statement

Hi,

sorry but how many rows are "7 core records" ?

What indexes do you have ?

What's the datatype of PUR_DATE ?

What is column PUR or is it a typo ?

Why do you use the TRUNC around your date columns ? You use trunc around the sysdate, that seems enough.

Please post the execution plan.

regards

Kay

So how many rows in that table fit the predicate of RES_Value = 'PUR'?

Is the index actually being used?

Should the index also allow you to filter on one of your date column as well or does that not reduce the number of blocks used by much?

Have you seen the actual execution plan being used? Where is the time going?

What is the relation between pur and pur_date? Is there one?

Maybe you'd benefit from filtering directly on your date columns rather than a function of them.

you should provide much more information at least your execution plan but since your query is too simple:

    FROM   ACC_REP
   WHERE   RES_Value = 'PUR'
           AND TRUNC (TO_DATE (PUR_DATE, 'dd/mm/yyyy')) >= TRUNC (SYSDATE - 1)
           AND TRUNC (TO_DATE (PUR, 'dd/mm/yyyy')) <= TRUNC (SYSDATE - (1-1))

this part is important for you. how many rows does this table has? how many of them has "PUR" value for res_value column? Are PUR_DATE and PUR columns varchar2 or date? when you run query above (with "select *", no group by) how many rows are returning? does your table statistics up to date ?

ps: AndrewSayer I wrote this post without seeing yours, you already ask whatever I asked.

Please find the query:

select sql_id, sql_text from v$sqlarea where upper(sql_text) like 'SELECT%V_D%SUBSTR%MAX%ACC_REP%' ;

from the sql_text identify your problematic query and then use:

select * from table(dbms_xplan.display_cursor('sql_id',null,'advanced')) ;

Then please post the execution plan here as an attachment, it will be easier to read as a text file then in the bad format that you would post here (maybe)

please see Andrew's and mine posts. rowcounts are important.

for pure_Date, you can write like this:

AND PUR_DATE >= TRUNC (SYSDATE - 1)

AND PUR_DATE <= TRUNC (SYSDATE - (1-1))

but it may not increase performance, do you have any index that contain, PUR_DATE and RES_VALUE ? please give some info about row counts.

Up to date statistics does not mean representative.

How many rows/blocks should your predicates return?

Don't use to_date on a timestamp.

No need to trunc your columns

Change predicates to

FROM   ACC_REP
   WHERE   RES_Value = 'PUR'
           AND PUR_DATE >=
                 TRUNC (SYSDATE - 1)
           AND pur_date <=
                 TRUNC (SYSDATE - (1-1))

And maybe index res_value, pur_date.

Share execution plan and the actual selectivities if that doesn't solve your problem.

-edit

It seems you editted your post after Id seen it and really the column is a varchar2. Were you not aware of this? Are you aware of how many problems this causes you? Can you fix the table so you are storing data in the correct types and therefore constrain the data properly and give the cost based optimizer the minimum amount of information it needs (stats would also be useful).

975791 wrote:

 

Hi All,

 

 

Thanks for your response......

 

What is column PUR or is it a typo ?

it is typo error.it is PUR_DATE,not PUR.

What's the datatype of PUR_DATE ?

PUR_DATE is varchar2(30)

 

What indexes do you have ?

Non unique index

 

Why do you use the TRUNC around your date columns ? You use trunc around the sysdate, that seems enough.

I couldn't understand it

 

does your table statistics up to date ?

Every week we gather stats.

 

Any Suggestions

 

Thanks,

First of all, the issue is storing date as VARCHAR2. Fix that and store the date as DATE.

Thanks,

Ann Edmund

The fact that PUR_DATE is a varchar2 makes this awkward. Predicates such as

AND PUR_DATE >= TRUNC (SYSDATE - 1)

may be disastrous, because they will be relying on implicit type casting. Don't even think about it.

You do need to say how many rows you expect from the combined predicate and how many for each of the predicates individually. Then you could consider creating a virtual column populated as to_date(pur_date, 'dd/mm/yyyy') and creating a composite index on RES_VALUE plus the virtual column (or perhaps the other way around) and adjusting your query to filter on the virtual column.

Note that you may have a problem with statistics. It would seem likely that analyzing once a week means that your stats know nothing of rows entered in the last day.

975791 wrote:

 

Hi All,

 

 

Thanks for your response......

 

What is column PUR or is it a typo ?

it is typo error.it is PUR_DATE,not PUR.

What's the datatype of PUR_DATE ?

PUR_DATE is varchar2(30)

 

What indexes do you have ?

Non unique index

 

Why do you use the TRUNC around your date columns ? You use trunc around the sysdate, that seems enough.

I couldn't understand it

 

does your table statistics up to date ?

Every week we gather stats.

 

Any Suggestions

 

Thanks,

As you've been told the design of this table is seriously flawed, using a VARCHAR2 datatype rather than using DATE.  The query relies on implicit date conversion, which I would't rely on for production data.  I blogged on such mistakes in design here:

https://dfitzjarrell.wordpress.com/2016/02/17/that-aint-right/

Additionally what you're doing is causing problems with the optimizer since Oracle can't know these values are dates; looking at an example using a VARCHAR2 column storing the date it produces incorrect results:

SQL> create table datetst(

  2  myid       number,

  3  mydt       varchar2(20));

Table created.

SQL>

SQL> begin

  2          for i in 1..100 loop

  3                  insert into datetst

  4                  values(i, to_char(sysdate+i, 'MM/DD/RRRR'));

  5          end loop;

  6

  7          commit;

  8  end;

  9  /

PL/SQL procedure successfully completed.

SQL>

SQL> select *

  2  from datetst

  3  where mydt <= sysdate +10;

where mydt <= sysdate +10

      *

ERROR at line 3:

ORA-01843: not a valid month

SQL>

SQL> select *

  2  from datetst

  3  where mydt <= to_char(sysdate +10, 'MM/DD/RRRR');

      MYID MYDT

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

         1 11/16/2016

         2 11/17/2016

         3 11/18/2016

         4 11/19/2016

         5 11/20/2016

         6 11/21/2016

         7 11/22/2016

         8 11/23/2016

         9 11/24/2016

        10 11/25/2016

        47 01/01/2017

      MYID MYDT

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

        48 01/02/2017

        49 01/03/2017

        50 01/04/2017

        51 01/05/2017

        52 01/06/2017

        53 01/07/2017

        54 01/08/2017

        55 01/09/2017

        56 01/10/2017

        57 01/11/2017

        58 01/12/2017

      MYID MYDT

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

        59 01/13/2017

        60 01/14/2017

        61 01/15/2017

        62 01/16/2017

        63 01/17/2017

        64 01/18/2017

        65 01/19/2017

        66 01/20/2017

        67 01/21/2017

        68 01/22/2017

        69 01/23/2017

      MYID MYDT

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

        70 01/24/2017

        71 01/25/2017

        72 01/26/2017

        73 01/27/2017

        74 01/28/2017

        75 01/29/2017

        76 01/30/2017

        77 01/31/2017

        78 02/01/2017

        79 02/02/2017

        80 02/03/2017

      MYID MYDT

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

        81 02/04/2017

        82 02/05/2017

        83 02/06/2017

        84 02/07/2017

        85 02/08/2017

        86 02/09/2017

        87 02/10/2017

        88 02/11/2017

        89 02/12/2017

        90 02/13/2017

        91 02/14/2017

      MYID MYDT

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

        92 02/15/2017

        93 02/16/2017

        94 02/17/2017

        95 02/18/2017

        96 02/19/2017

        97 02/20/2017

        98 02/21/2017

        99 02/22/2017

       100 02/23/2017

64 rows selected.

SQL>

The first error occurs because of the default NLS date format; an explicit conversion needed to take place to get any results from the query.  Notice also that this query returned 64 rows of data when the expected output should have been no more than 10 rows (given the data in the table).

Fixing that (and, yes, I know you can't fix this by changing the table definition) a different result is presented:

SQL> drop table datetst purge;

Table dropped.

SQL>

SQL> create table datetst(

  2  myid       number,

  3  mydt       date);

Table created.

SQL>

SQL> begin

  2          for i in 1..100 loop

  3                  insert into datetst

  4                  values(i, sysdate+i);

  5          end loop;

  6

  7          commit;

  8  end;

  9  /

PL/SQL procedure successfully completed.

SQL>

SQL> select *

  2  from datetst

  3  where mydt <= sysdate +10;

      MYID MYDT

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

         1 16-NOV-16

         2 17-NOV-16

         3 18-NOV-16

         4 19-NOV-16

         5 20-NOV-16

         6 21-NOV-16

         7 22-NOV-16

         8 23-NOV-16

         9 24-NOV-16

        10 25-NOV-16

10 rows selected.

SQL>

Now the 10 correct rows are returned and there wasn't any 'fiddling' with the query to get things to process. 

Part of your problem, I believe, goes back to the first part of this example where Oracle returned 64 rows when only 10 should have met the criteria.  I expect the poor decision of making the PUR_DATE column a VARCHAR2 datatype is causing far more records to be returned, records that do NOT actually meet the criteria, inflating the result set needlessly and causing more work for Oracle.

David Fitzjarrell