Trying to Remove table access full in oracle - sql

Question

i have a simple query
select round(sum(a.wt)) as a_wt
from db.abc a
where a.date is null
and a.col_no is not null
and a.pod_cd = '367'
and a.fant != a.rce
and I want to remove table access full.there are 3 index which are like these on following combination of column
col_no
col_no,date,fant,pyc
wagno,batno
what can be can be done to remove table access full.

Answer 1

One option would be creating a Function Based Index :
create index idx_date_col_pod on ABC (nvl("date",date'1900-01-01'), nvl(col_no,0), pod_cd);
and convert the query to :
select round(sum(wt)) as a_wt
from abc
where nvl("date",date'1900-01-01') = date'1900-01-01' -- matching means "date" column is null assuming there exists no records with this ancient date.
and nvl(col_no,0) != 0 -- non-matching means "col_no" column is not null
and pod_cd = 367
and fant != rce

Answer 2

Usually indexes are not indexing the null values, so the conditions like
where a.date is null
and a.col_no is not null
meaning just "don't use an index in order to get lines for these conditions"
However, there is an option in the create index statement allowing it to index null columns (starting from version 11 as far as I know)
create index abc_date_nulls on abc(date, 1); -- (xxx,1) is doing the trick
Thus you'll create an index that considers null values. This might be useful depending on selectivity of "date is null" condition.
Otherwise or in addition, I'd suggest you to check the selectivity for the condition "pod_cd = 367" and build an index on pod_cd column.
If you are sure the index will help and the database doesn't use it, you can force oracle to use an index using a hint
select /*+ index(index name) */ ... from ...
It is good for tests or for checking the impact indexes can provide but pleasepleaseplease be careful using them in production. Google the documentation and all the things about disadvantages for that approach. Don't tell anyone I told you to use hints on production

Answer 3

Indexes can be used to check for nulls, and to compare two columns against each other.
Setup:
create table abc
( dt date
, col_no number
, pod_cd varchar2(5)
, fant number
, rce number
, wt number )
nologging;
insert /*+ append */ into abc (dt, col_no, pod_cd, fant, rce, wt)
select case mod(rownum,3) when 0 then date '2018-12-31' + mod(rownum,1000) end
, case mod(rownum,7) when 0 then rownum end
, case mod(rownum,2) when 1 then mod(rownum,1000) end
, round(dbms_random.value) + 10
, round(dbms_random.value) + 10
, 1
from xmltable('1 to 10000000');
create index x1 on abc (pod_cd, dt);
create index x2 on abc (fant, rce);
Test for nulls:
select count(*) from abc a
where a.pod_cd = '367'
and a.dt is null;
COUNT(*)
----------
6667
1 row selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 2253536563
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 7 | 20 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 7 | | |
|* 2 | INDEX RANGE SCAN| X1 | 6667 | 46669 | 20 (0)| 00:00:01 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("A"."POD_CD"='367' AND "A"."DT" IS NULL)
The query was executed using index X1, without touching the table.
Test for fant != rce:
select count(*)
from abc a
where a.fant != a.rce;
COUNT(*)
----------
5000666
1 row selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 29151601
------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 6 | 6468 (1)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 6 | | |
|* 2 | INDEX FAST FULL SCAN| X2 | 5000K| 28M| 6468 (1)| 00:00:01 |
------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - filter("A"."FANT"<>"A"."RCE")
The query was executed using index X2, also without touching the table.
Test for the full query:
create index x3 on abc(pod_cd, dt, fant, rce, col_no, wt);
select round(sum(a.wt)) as a_wt
from abc a
where a.dt is null
and a.col_no is not null
and a.pod_cd = '367'
and a.fant != a.rce;
A_WT
----------
481
1 row selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 3828004431
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 16 | 28 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 16 | | |
|* 2 | INDEX RANGE SCAN| X3 | 476 | 7616 | 28 (0)| 00:00:01 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("A"."POD_CD"='367' AND "A"."DT" IS NULL)
filter("A"."COL_NO" IS NOT NULL AND "A"."FANT"<>"A"."RCE")
Full table scans aren't always terrible, though.
SQL> drop index x1;
Index dropped.
SQL> drop index x2;
Index dropped.
SQL> drop index x3;
Index dropped.
select round(sum(a.wt)) as a_wt
from abc a
where a.dt is null
and a.col_no is not null
and a.pod_cd = '367'
and a.fant != a.rce;
A_WT
----------
481
1 row selected.
Elapsed: 00:00:00.18
Execution Plan
----------------------------------------------------------
Plan hash value: 1045519631
---------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 16 | 8188 (1)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 16 | | |
|* 2 | TABLE ACCESS FULL| ABC | 476 | 7616 | 8188 (1)| 00:00:01 |
---------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - filter("A"."COL_NO" IS NOT NULL AND "A"."POD_CD"='367' AND
"A"."DT" IS NULL AND "A"."FANT"<>"A"."RCE")
Table ABC has 10 million rows, and the full scan took 0.18 seconds. This is in a VM on a 4 year old laptop.