db2 - Partitioning on Multiple Columns

Posted 2022-10-29 t0nsha

Partitioning on Multiple Columns

Multiple columns can be specified in the PARTITION BY clause. This is similar to defining multiple columns in an index key:

CREATE TABLE SALES (YEAR INT, MONTH INT)
PARTITION BY RANGE(YEAR, MONTH)
  (
  STARTING (2007,1)  ENDING (2007,3),
  STARTING (2007,4)  ENDING (2007,6),
  STARTING (2007,7)  ENDING (2007,9),
  STARTING (2007,10) ENDING (2007,12),
  STARTING (2008,1)  ENDING (2008,3)
  )

Note that multiple columns are not multiple dimensions. The ranges cannot overlap or else DB2 will raise an error condition:

CREATE TABLE NEIGHBORHOODS (STREET INT, AVENUE INT )
PARTITION BY RANGE(STREET, AVENUE)
  (
  STARTING (1,1)  ENDING (10,10),
  STARTING (1,11) ENDING (10,20)
) ;
SQL0636N  Range specified for data partition "PART0" is not valid.  Reason code = "10".  SQLSTATE=56016

The reason for the error is due to the overlapping ranges between the two partition specifications. The first range goes from (1,1) to (10,20) while the second goes from (1,11) to (10,20). While this would appear to be reasonable, the problem is with the way ranges are computed. Ranges are linear so when the second range ends at (1,20), it starts back over at (2,1) instead of (2,11) as one might expect. If the data needs to be organized by grids or cubes the Multidimensional Clustering (MDC) feature should be used instead.

FROM: http://www.toadworld.com/platforms/ibmdb2/w/wiki/7352.partitioning-on-multiple-columns