Query tuning : you can do it yourself

Question

Hello,
This thread is not a question.
How tune my query ?
My query didn't use indexes why ?
I have indexes on table, stats are up-to-date why Oracle does a full scan on table ?
What are the hint which I can use ?
etc...
I would like try to answer to many such questions posted in this forum about query tuning.
And explain that tuning is not always complicated, is not always reserved to some consultants, is not always solved by hints usage, and not require to buy some books which would give some magic solutions.
By this thread, I would explain to people which have some query performance issues, that the solution is maybe here, in front of their eyes, inside the query itself. The solution may come from their own side, and that they can be happy to solve such question themself.
I'll develop here below a case from a real word situation encountered some time ago.

Answer 1

At this point, remove this huge and nightmarish hint. We'll use it only if no other ways may found.
First, a look into the data model to see if the joins may be rewrite differently.
Original joins :
    dw.AIRPORT_ROUTING_ID = ar.AIRPORT_ROUTING_ID and
    dw.COUNTRY_ROUTING_ID = cor.COUNTRY_ROUTING_ID and
    dw.REGION_ROUTING_ID = rr.REGION_ROUTING_ID and
    dw.CITY_ROUTING_ID  = cr.CITY_ROUTING_ID andAnd indeed, the join may be rewrite as below :
    and dw.AIRPORT_ROUTING_ID = ar.AIRPORT_ROUTING_ID
    and ar.CITY_ROUTING_ID    = cr.CITY_ROUTING_ID
    and cr.COUNTRY_ROUTING_ID = cor.COUNTRY_ROUTING_ID 
    and cor.REGION_ROUTING_ID = rr.REGION_ROUTING_IDNote the aliases were change here. Is it complicated to try to rewrite query like that ?
Secondly, we will try to see how avoid or, at least, workaround the all OR conditions. It would be well if we could merge all the FROM (bind variable :a) in one column, same for all TO (bind variable :b), unfortunately, the data model cannot be change (of course). Ok, we will try to work with a temporary table by combining all the possiblities between the two conditions.
create global temporary table REGION_ROUTING_TMP
on commit preserve rows
as
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_FROM as REGION_FROM, REGION_STOP1 as REGION_TO
from   REGION_ROUTING
where  1=2;And an index based on the two mains criteria of the query may help :
create index IDX_REGION_ROUTING_TMP on REGION_ROUTING_TMP(REGION_FROM,REGION_TO);Then we'll populate table :
insert into REGION_ROUTING_TMP
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_FROM as REGION_FROM, REGION_STOP1 as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_FROM as REGION_FROM, REGION_STOP2 as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_FROM as REGION_FROM, REGION_STOP3 as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_FROM as REGION_FROM, REGION_TO as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_STOP1 as REGION_FROM, REGION_STOP2 as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_STOP1 as REGION_FROM, REGION_STOP3 as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_STOP1 as REGION_FROM, REGION_TO as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_STOP2 as REGION_FROM, REGION_STOP3 as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_STOP2 as REGION_FROM, REGION_TO as REGION_TO
from   REGION_ROUTING
union
select REGION_ROUTING_NAMES, REGION_ROUTING_ID, REGION_STOP3 as REGION_FROM, REGION_TO as REGION_TO
from   REGION_ROUTING;The insert take near 9 seconds, which is a reasonnable elapse time.
At this point, the query become :
Select
   ap.AIRPORT_PAIR,  
   dw.DELIVERY_PERIOD,
   dw.TOTAL_PAX As Demand,
   od.GLOBAL_TOTAL_PAX as Total_OD_Demand,
   dw.REPORTED_PAX As ReportedPax,
   ap.AIRPORT_FROM,
   ap.AIRPORT_TO,
   ap.CITY_FROM,
   ap.CITY_TO,
   ap.COUNTRY_FROM_NAME,
   ap.COUNTRY_TO_NAME,
   ap.REGION_FROM_NAME,
   ap.REGION_TO_NAME,
   ar.AIRPORT_ROUTING,
   cr.CITY_ROUTING,
   cor.COUNTRY_ROUTING_NAMES,
   rr.REGION_ROUTING_NAMES,
   ar.NUMBER_OF_STOP,
   dw.PERCENT_OF_TOTAL_PAX_VS_OD,
   dw.AIRLINE1,
   dw.AIRLINE2,
   dw.AIRLINE3,
   dw.AIRLINE4,
   dw.NUMBER_OF_AIRLINE,
   dw.AIRCRAFTCLASS_NAME,
   dw.FARE
From
   REGION_ROUTING_TMP rr, -- here we use now the temporary table instead of REGION_ROUTING table
   DETAILED_FLIGHTS dw,
   AIRPORT_PAIR ap,
   AIRPORT_ROUTING ar,
   CITY_ROUTING cr,
   COUNTRY_ROUTING cor,
   OD od
where
        dw.DELIVERY_PERIOD  in ('2006-09','2006-08','2006-07')
and dw.AIRPORT_PAIR_ID    = ap.AIRPORT_PAIR_ID
and dw.OD_ID              = od.OD_ID
and dw.AIRPORT_ROUTING_ID = ar.AIRPORT_ROUTING_ID
and ar.CITY_ROUTING_ID    = cr.CITY_ROUTING_ID
and     cr.COUNTRY_ROUTING_ID = cor.COUNTRY_ROUTING_ID 
and cor.REGION_ROUTING_ID = rr.REGION_ROUTING_ID
and rr.REGION_FROM        = '7'  |
and rr.REGION_TO          = '8'  |-> these two lines replace all the OR conditions;And explain plan :
| Id  | Operation                             |  Name                          | Rows  | Bytes | Cost  | Pstart| Pstop |
|   0 | SELECT STATEMENT                      |                                |   469 |   174K|  3462 |       |       |
|   1 |  NESTED LOOPS                         |                                |   469 |   174K|  3462 |       |       |
|   2 |   NESTED LOOPS                        |                                |   458 |   166K|  3004 |       |       |
|   3 |    NESTED LOOPS                       |                                |   454 |   127K|  2096 |       |       |
|   4 |     NESTED LOOPS                      |                                |   258 | 58308 |   548 |       |       |
|   5 |      NESTED LOOPS                     |                                |   246 | 48216 |    56 |       |       |
|   6 |       NESTED LOOPS                    |                                |    26 |  4420 |     4 |       |       |
|   7 |        TABLE ACCESS BY INDEX ROWID    | REGION_ROUTING_TMP             |     1 |   123 |   
|*  8 |         INDEX RANGE SCAN              | IDX_REGION_ROUTING_TMP         |     1 |       |     1 |       | 
|   9 |        TABLE ACCESS BY INDEX ROWID    | COUNTRY_ROUTING                |    32 |  1504 |     2
|* 10 |         INDEX RANGE SCAN              | IX9_COUNTRY_ROUTING            |    32 |       |     1 |       | 
|  11 |       TABLE ACCESS BY INDEX ROWID     | CITY_ROUTING                   |     9 |   234 |     2 | 
|* 12 |        INDEX RANGE SCAN               | IX9_CITY_ROUTING               |    12 |       |     1 |    
|  13 |      TABLE ACCESS BY INDEX ROWID      | AIRPORT_ROUTING                |     1 |    30 |     2
|* 14 |       INDEX RANGE SCAN                | IX10_AIRPORT_ROUTING           |     1 |       |     1 |   
|* 15 |     TABLE ACCESS BY GLOBAL INDEX ROWID| DETAILED_FLIGHTS               |     2 |   122 |   
|* 16 |      INDEX RANGE SCAN                 | IX3_DETAILED_FLIGHTS           |     9 |       |     2 |    
|  17 |    TABLE ACCESS BY INDEX ROWID        | AIRPORT_PAIR                   |     1 |    85 |     2 |  
|* 18 |     INDEX RANGE SCAN                  | IX1_AIRPORT_PAIR               |     1 |       |     1 |       |
|  19 |   TABLE ACCESS BY INDEX ROWID         | OD                             |     1 |    10 |     1 |       |   
|* 20 |    INDEX UNIQUE SCAN                  | PK_OD                          |     1 |       |       |       |       |
----------------------------------------------------------------------------------------------------Ouf, this sound like better, isn't it ? Was it complicated to arrive as such result ?
Ok, now launch the query and see the elapse time :
104103 rows selected.
Elapsed: 00:09:00.12
Statistics
          0  recursive calls
          0  db block gets
    1214845  consistent gets
     272481  physical reads
          0  redo size
   11164355  bytes sent via SQL*Net to client
     257032  bytes received via SQL*Net from client
       6942  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
     104103  rows processedWoaw, developer is now very happy, less than 10 minutes instead of one hour... and without modify anything else in database and server settings, without add indexes, without add any hints...
Ok, now the problem is in the number of returned lines. Have you see the number of return client/server ? We'll now to try to reduce this one. Increasing the number of line returned by fetch :
set arraysize 5000
--run the query
104103 rows selected.
Elapsed: 00:02:25.49
Statistics
          0  recursive calls
          0  db block gets
    1150893  consistent gets
     271432  physical reads
          0  redo size
   11018641  bytes sent via SQL*Net to client
       1013  bytes received via SQL*Net from client
         22  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
     104103  rows processedOk, the number of retunr client/server is now low, and elapse time is ok user.