Select Most recent Occurance

Question

Hi Friends,
I have a strange requirement where in we need to select the most recent occurance of a record based on 2 columns, consider the below sample data:
C1     C2     C3
1234     0     abc
1234     1     abc
1234     0     def
1234     1     def
1345     20     fgh
1345     30     fgh
1345     20     ijk
1345     30     ijk
now i need to write a select statement which returns the following result
C1     C2     C3
1234     0     def
1234     1     def
1345     20     ijk
1345     30     ijk
we need to pick data from the recent occurance of the columns C1 and C2.
Please help me. Do let me know in case you need any thing.
Thanks.

Answer 1

Of course rowid not near always guarantees that:
SQL> drop table t;
Table dropped
SQL> create table t as (
  2  select 1234 C1, 0  C2, 'abc' C3 from dual)
  3  /
Table created
SQL> alter table t minimize records_per_block;
Table altered
SQL> insert into t
  2  select  1234, 1, 'abc' from dual union all
  3  select 1234, 0, 'def' from dual union all
  4  select 1234, 1, 'def' from dual union all
  5  select 1345, 20, 'fgh' from dual union all
  6  select 1345, 30, 'fgh' from dual union all
  7  select 1345, 20, 'ijk' from dual union all
  8  select 1345, 30, 'ijk' from dual
  9  /
7 rows inserted
SQL> select * from t;
        C1         C2 C3
      1234          0 abc
      1345         30 ijk
      1234          1 abc
      1234          0 def
      1234          1 def
      1345         20 fgh
      1345         30 fgh
      1345         20 ijk
8 rows selected
SQL> select * from t where rowid in(
  2     select max(rowid) over(partition by c1,c2 order by null) from t
  3  );
        C1         C2 C3
      1234          0 def
      1234          1 def
      1345         30 fgh
      1345         20 ijk
SQL> There is no guarantee that Oracle will insert your rows with continuously growing rowids especially if you have blocks with deleted rows, ASSM tablespaces etc. The only guarantee is to use some sequence column, because as I understand rows are inserted from flat file and therefore datetime also may be inappropriate due to the fact that rows can be inserted faster than datetime increases.
See also this thread where a person had similar problem Row is changing position - ON INSERT
Gints Plivna
http://www.gplivna.eu