How to create a range partitioned table

Using interval partitioning:

A simple example of an Interval Range Partitioned table:

CREATE TABLE DOCUMENTS
(
  DOCNAME          VARCHAR2(100 BYTE),
  TRACKING#        VARCHAR2(32 BYTE),
  ESTABLISHED_DT   DATE
)
PARTITION BY RANGE (ESTABLISHED_DT)
INTERVAL
(
   NUMTOYMINTERVAL(1,'MONTH')
)
(
PARTITION P_INIT VALUES LESS THAN (TO_DATE('2014-03','YYYY-MM') )
)
TABLESPACE USERS;

If you use INTERVAL partitioning, you are not allowed to specify an overflow partition.

Without internval partitioning:

Before Oracle 11g, this was the only way to create a RANGE partitioned table:
You had to manually define your partitions, and remember to manually add new ones if you expected new partition key values to be inserted.

Here I am using a column of type NUMBER to implement range partitioning.
I am defining an overflow partition to catch values outside the range of expected values:

CREATE TABLE mytable
(
  ID                      NUMBER(18)           NOT NULL,
  f_name                  NUMBER(18)           NOT NULL,
  l_name                  NUMBER(18)           NOT NULL,
  created_year            NUMBER(4)            NOT NULL
   ....
)
PARTITION BY RANGE (created_year)
(
PARTITION P_2003 values less than (2004) TABLESPACE 2003_DATA,
PARTITION P_2004 values less than (2005) TABLESPACE 2004_DATA,
PARTITION P_2005 values less than (2006) TABLESPACE 2005_DATA,
PARTITION P_DEFAULT values less than (MAXVALUE)
);

How to speed up statistics gathering on a large table - inter object parallelism

When you use parallelism to analyze a specific object, Oracle refer to this as "Inter Object Parallelism".
In short, the Degree Of Parallelism (DOP) is decided based on the object's DEGREE setting, OR the DEGREE Directive in the dbms_stats.gather_table_stats procedure.

In my example, the partitioned table HR.SALES was created with DEGREE=1, which is the default.

I then collected statistics on it, like this:

BEGIN
  SYS.DBMS_STATS.GATHER_TABLE_STATS (
     OwnName        => 'HR'
    ,TabName        => 'SALES'
    ,Degree            => DBMS_STATS.DEFAULT_DEGREE 
);
END;
/

When dbms_stats.gather_table_stats was executed with DEGREE=DBMS_STATS.DEFAULT_DEGREE, Oracle used a DOP of 8, which is calculated using the formula

PARALLEL_THREADS_PER_CPU (2) * CPU_COUNT (4)

If you look in v$sql, this Call to dbms_stats triggers a parallel SQL:

/* SQL Analyze(0) */
select /*+ full(t) parallel(t,8) parallel_index(t,8) dbms_stats cursor_sharing_exact use_weak_name_resl dynamic_sampling(0) no_monitoring xmlindex_sel_idx_tbl no_substrb_pad */

I found that for a table sized 100G, the statistics gathering dropped from 3 hours to 47 minutes using the DEGREE directive.

Note that Oracle will still collect statistics one partition at a time:

“You should note that setting the DEGREE for a partitioned table means that multiple parallel sever processes will be used to gather statistics on each partition but the statistics will not be gathered concurrently on the different partitions. Statistics will be gathered on each partition one after the other”.

Sources: Whitepaper "Best Practices for Gathering Optimizer Statistics with Oracle Database 12c"
Oracles Documentation about dbms_stats

Oracle 12c feature: Online statistics gathering

Oracle 12c introduced the concept of "online statitistics gathering".

The white paper "Best Practices for Gathering Optimizer Statistics with Oracle Database 12c" released in June 2013, explains:


"In Oracle Database 12c, online statistics gathering "piggybacks" statistics gather as part of a direct-path data loading operation"

and

Gathering statistics as part of the data loading operation, means no additional full data scan is required to have statistics available immediately after the data is loaded.”

and

"online statistics gathering was designed to have a minimal impact on the performance of a direct path load operation it can only occur when data is being loaded into an empty object"

This can be a huge time saver in database where regular heap-organized tables are loaded and truncated as a part of their ETL Processing.

Note that some restrictions apply:

1. Histograms or index statistics are not gathered as a part of online statistics gathering, since they require additional data scans

2. Online statistics gathering will only follow a Direct-path load:

   * CTAS - Create Table As Select
   * IAS - Insert As Select
   * INSERT INTO /* +APPEND */

3. The base tables which are loaded must be empty

Any negative consequence of the first restriction can be mitigated by collecting historgram data on the table *after* the online statistics have been collected. Oracle does this without re-gathering the base column statistics:

exec dbms_stats.gather_table_stats(OwnName => 'SCOTT',TabName => 'DEPT', options=>'GATHER AUTO');

The easiest way to check if the statistics in a column has been gathered via the Online statisitcs feature, is to check the "histogram" and the "notes" column in the user|dba|all _tab_col_statistics view:

set lines 200
col column_name format a30
col notes format a30
select column_name,num_distinct,num_nulls,histogram,notes
from dba_tab_col_statistics
where table_name='DEPT';

If output in the histogram column says "NONE" and the notes columns says "STATS_ON_LOAD", Online statistics gathering has indeed been utilized.

If the output of the same columns says "HYBRID" and "HISTOGRAM_ONLY", it indicates that histograms were gathered without regathering basic column statistics.

How to solve ORA-14323: cannot add partition when DEFAULT partition exists

You have a partitioned table with an overflow partition, and you try to add another partition to it:

ALTER TABLE RECEIVED_DOCUMENTS 
ADD PARTITION EES_COUNTRIES VALUES('NORWAY','ICELAND')

Result:

ERROR at line 1:
ORA-14323: cannot add partition when DEFAULT partition exists

Solution: split the overflow partition:

ALTER TABLE RECEIVED_DOCUMENTS 
SPLIT PARTITION "OTHERS" 
VALUES ('NORWAY','ICELAND')
INTO ( PARTITION EES_COUNTRIES,
       PARTITION OTHERS
       )
UPDATE INDEXES;

How to split an overflow partition using ALTER TABLE .... SPLIT PARTITION

For a range partitioned table:

ALTER TABLE SCOTT.EMPLOYEES
SPLIT PARTITION START_DATE_MAX AT (TO_DATE('2014-03-01', 'YYYY-MM-DD')) 
INTO (
      PARTITION START_DATE_FEB_2014 TABLESPACE ACTIVE_DATA,
      PARTITION START_DATE_MAX TABLESPACE TOOLS
     );

For a list-partitioned table:

ALTER TABLE received_documents
SPLIT PARTITION OTHERS
VALUES ('EAST','WEST','NORTH','SOUTH')
INTO (PARTITION NON_EU_CITIZEN,
      PARTITION OTHERS
) UPDATE INDEXES PARALLEL 8;

See this post for an example of splitting a default subpartition.

How to create a global temporary table

An example of a Global Temporary Table (GTT) would be:

CREATE GLOBAL TEMPORARY TABLE INCOMING_DATA (
  ROW_ID                      NUMBER(10) NOT NULL,
  SSN                         NUMBER(11),
  ENTRY_DATE                  DATE NOT NULL,
  HANDLED_BY                  VARCHAR2(30 BYTE) NOT NULL,
  CONTENT_CODE                VARCHAR2(4 BYTE),
  MESSAGE                     CLOB
 )
ON COMMIT DELETE ROWS
;

ON COMMIT DELETE ROWS indicates that Oracle will purge the rows after an ended transaction (after a COMMIT, ROLLBACK or an implicit COMMIT).
Instead of ON COMMIT DELETE ROWS you could also use ON DELETE PRESERVE ROWS, which will remove the rows at the end of the user's session.

You can add indexes, triggers and views on a GTT. You can even truncate a temporary table, and it will only affect the curren session's rows, leaving other users' rows intact.


With Oracle 12c came the ability for temporary tables to store its UNDO segments in temporary tablespaces, rather than conventional tablespaces, thus removing the need to generate REDO.

To enable this for your session, use:

ALTER SESSION SET TEMP_UNDO_ENABLED = TRUE;

before creating the GTT.

Sources:

Oracle-Base: article about GTT and another one explaining the 12c new feature "temporary undo"

Oracle Documentation: Overview of GTT
Check my note on private temporary tables, available from Oracle 18c.

Create a globally partitioned hash index

Partitioned table layout:

CREATE TABLE SCOTT.ARCHIVED_NOTES
(
  NOTE_ID                     NUMBER(10)        NOT NULL,
  PPT_ID                      NUMBER(10),
  REGISTRED_DATE              DATE              NOT NULL,
  EDITED_DATE                 DATE,
  ALTERED_BY                  VARCHAR2(30 BYTE) NOT NULL,
  ALTERED_DATE                DATE              NOT NULL,
  CONTENT_CODE                VARCHAR2(4 BYTE),
  NOTEFORMAT                  VARCHAR2(1 BYTE)
)
TABLESPACE USERS
PARTITION BY RANGE (NOTE_ID)
(  
  PARTITION ARCHIVED20 VALUES LESS THAN (21000000),  
  PARTITION ARCHIVED21 VALUES LESS THAN (22000000),
  PARTITION ARCHIVED22 VALUES LESS THAN (23000000)
)
;

Create a global hash-partitioned index. Notice the GLOBAL keyword:

CREATE INDEX ARCHIVED_NOTES_IDX2 ON ARCHIVED_NOTES (REGISTRED_DATE)
  GLOBAL PARTITION BY HASH (REGISTRED_DATE)
  PARTITIONS 3;

Verify its existence:

SELECT INDEX_NAME,PARTITION_NAME 
FROM USER_IND_PARTITIONS 
WHERE INDEX_NAME = (SELECT INDEX_NAME FROM USER_INDEXES WHERE TABLE_NAME='ARCHIVED_NOTES'); 

INDEX_NAME	PARTITION_NAME
ARCHIVED_NOTES_IDX2	SYS_P441
ARCHIVED_NOTES_IDX2	SYS_P442
ARCHIVED_NOTES_IDX2	SYS_P443

Let's take a deeper look at the index properties:

SELECT C.TABLE_NAME,I.INDEX_NAME,I.PARTITIONING_TYPE, I.SUBPARTITIONING_TYPE, I.LOCALITY,I.ALIGNMENT,C.COLUMN_NAME,C.COLUMN_POSITION
FROM USER_PART_INDEXES I JOIN USER_IND_COLUMNS C
ON (I.INDEX_NAME = C.INDEX_NAME)
WHERE I.TABLE_NAME IN (SELECT TABLE_NAME FROM USER_TABLES WHERE PARTITIONED='YES')
ORDER BY 1,2,7;

TABLE_NAME	INDEX_NAME	PARTITIONING_TYPE	SUBPARTITIONING_TYPE	LOCALITY	ALIGNMENT	COLUMN_NAME	COLUMN_POSITION
ARCHIVED_NOTES	ARCHIVED_NOTES_IDX2	HASH	NONE	GLOBAL	PREFIXED	REGISTRED_DATE	1

For global partitioned indexes, the "alignment" column will always show the index as "prefixed". This is the only supported index type for global partitioned indexes.