Oracle DB 10g - Count records from a subquery is to slow. (after tuning)

Question

Hi everybody:
I have the following problem.
1.- I have a transactional table with 11 millions of records.
2.- The user asked for a few reports, to know some sumarized data for business decisions.
3.- Initialy the query was developed in a development environment, with a subset smaller than the actual data mentioned in label 1.
4.- When the report was delivered to the end user, the report never return data.
5.- at this point, we performed tuning, adding indexes, re-writing the query, using hints, etc. and the following scenario is ocurring:
a) the query without the count, before the tuning was taking about 3 to 5 minutes, and returned aproximatelly 332,000 records-.
b) the numer of records was counted and this query takes aproximatelly 15 to 23 minutes.
c) after the tuning, the raw data, returns in 1 second, we used some b-tree indexes, some FBI (because the report needs to filter by to_char functions).
that time is ok for us, 1 second is a great time in comparison with the 3 to five minutes.
d) the funny thing, is that when we add a simple count(1), count(x) or wathever flavor of count, the count takes about 3 minutes to return, with the 332,00
records. is better than de 15 minutes of course, but we dont need count(1), we need to use group by, order by, etc, and will increase the time of query.
6.- Another thing is happening, if I use count(1) with a transactional table with 600,000 records, without filtering, the count returns in les than 2 seconds. Even if the data is more than the result of my query defined in label 5.c, and that query returns in 1 second.
Please help me with this, I know that maybe is something that i'm not considering on tuning. Or if there is a way to run this count query faster, please let me know.
This is the query:
WITH historial_depurado_v AS (
select serie, identificador,
       cc_form_cc_tifo_tipo_forma
      ,cc_form_serie
      ,cc_form_folio
      ,estatus_nuevo
      ,to_char(fecha,'MM') Mes
      ,to_char(fecha,'YYYY') Anio  -- the table has a FBI for this.
      ,get_recaudacion_x_forma (hifo.CC_FORM_CC_TIFO_TIPO_FORMA
                               ,hifo.CC_FORM_SERIE
                               ,hifo.CC_FORM_FOLIO) Recaudacion  -- function for description.
from  cc_historial_formas hifo
where not exists (select 1
                     from ve_tipos_placas tipl
                    where tipl.cc_tifo_tipo_forma = hifo.cc_form_cc_tifo_tipo_forma)
SELECT serie
  FROM historial_depurado_v hide
WHERE Anio = '2009'  -- using the function based index.
   AND Mes = '01'
   AND Estatus_nuevo = 'UT'
-- returns in 1 second, but if I use count(serie) takes 3 to 5 minutes, and still is not complete, I need to use group by some fields returned by the firs subset.
-- if I count a table with more records, returns in 2 seconds.

Answer 1

alopez wrote:
WHERE Anio = '2009'  -- using the function based index.
AND Mes = '01'
AND Estatus_nuevo = 'UT'
Also, i'm not sure if your internal comment is correct, but if you have a function based index on any of those columns, it will ONLY work when you apply the exact function as defined in the FBI (you aren't using any functions on ANY of your columns).
An example.
create table use_the_fbi
   column1  date
insert into use_the_fbi
select
   trunc(sysdate, 'dd') - level
from dual connect by level <= 10000;
create index use_the_fbi_FBI on use_the_fbi (to_char(column1, 'YYYYMM') );
exec dbms_stats.gather_table_stats( user, 'USE_THE_FBI', cascade=> true, method_opt=> 'for all hidden columns');Let's try without querying in the manner in which we defined the index.
explain plan for
select *
from use_the_fbi
  4  where column1 = '201006';
Explained.
Elapsed: 00:00:00.01
TUBBY_TUBBZ?@xplain
PLAN_TABLE_OUTPUT
Plan hash value: 1722805582
| Id  | Operation         | Name        | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT  |             |   100 |   900 |     7   (0)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| USE_THE_FBI |   100 |   900 |     7   (0)| 00:00:01 |
Query Block Name / Object Alias (identified by operation id):
   1 - SEL$1 / USE_THE_FBI@SEL$1
Predicate Information (identified by operation id):
   1 - filter("COLUMN1"='201006')
Column Projection Information (identified by operation id):
   1 - "COLUMN1"[DATE,7]
23 rows selected.
Elapsed: 00:00:00.02Sad Christmas .. no index love.
Now, query as we defined the index
explain plan for
select *
from use_the_fbi
where to_char(column1, 'YYYYMM') = '201006';
Explained.
Elapsed: 00:00:00.01
TUBBY_TUBBZ?TUBBY_TUBBZ?@xplain
PLAN_TABLE_OUTPUT
Plan hash value: 268331390
| Id  | Operation                   | Name            | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT            |                 |    30 |   450 |     2   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| USE_THE_FBI     |    30 |   450 |     2   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | USE_THE_FBI_FBI |    30 |       |     1   (0)| 00:00:01 |
Query Block Name / Object Alias (identified by operation id):
   1 - SEL$1 / USE_THE_FBI@SEL$1
   2 - SEL$1 / USE_THE_FBI@SEL$1
Predicate Information (identified by operation id):
   2 - access(TO_CHAR(INTERNAL_FUNCTION("COLUMN1"),'YYYYMM')='201006')
Column Projection Information (identified by operation id):
   1 - "USE_THE_FBI"."COLUMN1"[DATE,7]
   2 - "USE_THE_FBI".ROWID[ROWID,10], TO_CHAR(INTERNAL_FUNCTION("COLUMN1"),'YYYYMM')[VA
       RCHAR2,6]
27 rows selected.
Elapsed: 00:00:00.03
TUBBY_TUBBZ?Happy times.