--+APPEND Hint -- advantages and disadvantages in OLTP system

Question

I have a process in with an insert statement takes a lot of time. When I modify the insert statement to use --+ APPEND hint the insert time reduces a lot. There are a lot of indexes on this table and the table and indexes are in logging mode.
Can someone please clarify is using +APPEND hint will have any disadvantage in an OLTP system . Will any data be lost in case of failure?
Thanks a lot!!

Answer 1

Jim,
Append means two things: 1) don't generate redo if possible, 2) don't run around looking through freelists for existing blocks into which a stray row or three could fit, just insert the rows into new empty blocks.
The OP stated that the table is in logging mode. This means that the insert/append will be logged and hence the transaction is recoverable. It also means that append doesn't get rid of redo generation, it only gets rid of the freelist overhead.
I'll create a table that occupies multiple blocks:
SQL> create table t as
  2  select * from user_objects;
Table created.
SQL> insert into t
  2  select * from t;
58 rows created.
SQL> /
116 rows created.
SQL> /
232 rows created.
SQL> /
464 rows created.
SQL> select rowid_block, count(*) cnt
  2    from (
  3  select dbms_rowid.rowid_block_number(rowid) AS rowid_block
  4    from t
  5         )
  6   group by rowid_block;
ROWID_BLOCK        CNT
      32542         58
      32543        183
      32544        183
      32545        183
      32546        183
      32547        138
6 rows selected.Then I'll empty most of the rows out of those blocks:
SQL> delete from t
  2   where object_id != 32217;
912 rows deleted.
SQL> select rowid_block, count(*) cnt
  2    from (
  3  select dbms_rowid.rowid_block_number(rowid) AS rowid_block
  4    from t
  5         )
  6   group by rowid_block;
ROWID_BLOCK        CNT
      32542          1
      32543          4
      32544          3
      32545          3
      32546          3
      32547          2
6 rows selected.Then I'll perform a "normal" bulk insert and we'll see the data added to an existing block:
SQL> insert into t
  2  select * from t;
16 rows created.
SQL> select rowid_block, count(*) cnt
  2    from (
  3  select dbms_rowid.rowid_block_number(rowid) AS rowid_block
  4    from t
  5         )
  6   group by rowid_block;
ROWID_BLOCK        CNT
      32542          1
      32543          4
      32544          3
      32545          3
      32546          3
      32547         18
6 rows selected.However, a bulk insert/append will use a new block -- bypassing the freelist:
SQL> insert /*+ append */ into t
  2  select * from t;
32 rows created.
SQL> commit;
Commit complete.
SQL> select rowid_block, count(*) cnt
  2    from (
  3  select dbms_rowid.rowid_block_number(rowid) AS rowid_block
  4    from t
  5         )
  6   group by rowid_block;
ROWID_BLOCK        CNT
       2133         32
      32542          1
      32543          4
      32544          3
      32545          3
      32546          3
      32547         18
7 rows selected.Scott