ALTER TABLE PARTITION

Function

ALTER TABLE PARTITION modifies table partitioning, including adding, deleting, splitting, merging partitions, and modifying partition attributes.

Precautions

• The name of the added partition must be different from names of existing partitions in the partitioned table.
• For a range partitioned table, the boundary value of the added partition must be the same type as the partition key of the partitioned table. The key value of the added partition must exceed the upper limit of the last partition.
• For a list partitioned table, if the DEFAULT partition has been defined, no new partition can be added.
• Unless otherwise specified, the syntax of range partitioned tables is the same as that of column-store partitioned tables.
• If the number of partitions in the target partitioned table has reached the maximum (32767), partitions cannot be added.

• If a partitioned table has only one partition, the partition cannot be deleted.
• Use PARTITION FOR() to choose partitions. The number of specified values in the brackets should be the same as the column number in customized partition, and they must be consistent.
• The Value partitioned table does not support the Alter Partition operation.
• For OBS multi-temperature tables:
  • The tablespace of a partitioned table cannot be an OBS tablespace during the MOVE, EXCHANGE, MERGE, and SPLIT operations.
  • When an ALTER statement is executed, the cold and hot data attributes in the partitions cannot be changed, that is, data in the cold partition should still be in the cold partition after a data operation, and hot partition data should be in the hot partition. Cold partition data cannot be migrated to the local tablespace.
  • Only the default tablespace is supported for cold partitions.
  • Cold and hot partitions cannot be merged.
  • Cold partition switching is not supported for the EXCHANGE operation.

Syntax

• Modify the syntax of the table partition.

  1 2	ALTER TABLE [ IF EXISTS ] { table_name [*] | ONLY table_name | ONLY ( table_name )} action [, ... ];

  action indicates the following clauses for maintaining partitions. For the partition continuity when multiple clauses are used for partition maintenance, GaussDB(DWS) does DROP PARTITION and then ADD PARTITION, and finally runs the rest clauses in sequence.

  1 2 3 4 5 6 7 8 9

  move_clause | exchange_clause | row_clause | merge_clause | modify_clause | split_clause | add_clause | drop_clause |

  • The exchange_clause syntax is used to move the data from a general table to a specified partition.

    1 2 3

    EXCHANGE PARTITION { ( partition_name ) | FOR ( partition_value [, ...] ) } WITH TABLE {[ ONLY ] ordinary_table_name | ordinary_table_name * | ONLY ( ordinary_table_name )} [ { WITH | WITHOUT } VALIDATION ] [ VERBOSE ]

    The ordinary table and the partitioned table whose data is to be exchanged must meet the following requirements:

    • The number of columns of the ordinary table is the same as that of the partitioned table, and their information should be consistent, including the column name, data type, constraint, collation, storage parameter, compression, and data type of a deleted column.
    • The compression information of the ordinary table and partitioned table should be consistent.
    • The distribution column information of the ordinary table and the partitioned table should be consistent.
    • The number and information of indexes of the ordinary table and the partitioned table should be consistent.
    • The number and information of constraints of the ordinary table and the partitioned table should be consistent.
    • The ordinary table cannot be a temporary table or unlogged table.
    • The ordinary table and the partitioned table must be in the same logical cluster or node group.
    • If other columns following the last valid column in the partitioned table are deleted and the deleted columns are not considered, the partitioned table can be exchanged with the ordinary table as long as the columns of the two tables are the same.
    • The table-level parameter colversion of a column-store ordinary table must be the same as that of a column-store partitioned table. Partition swap between colversion2.0 and colversion1.0 is not allowed.

    When the execution is complete, the data and tablespace of the ordinary table and the partitioned table are exchanged. In this case, statistics about the ordinary table and the partitioned table become unreliable. Both tables should be analyzed again.

  • The syntax of row_clause is used to set the row movement switch of a partitioned table.

    1	{ ENABLE | DISABLE } ROW MOVEMENT

  • The merge_clause syntax is used to merge partitions into one.

    1	MERGE PARTITIONS { partition_name } [, ...] INTO PARTITION partition_name

    

    • The partition before the keyword INTO is called the source partition, and the partition after the INTO is called the target partition.
    • The number of source partitions ranges from 2 to 32.
    • The source partition name must be unique.
    • The source partition cannot have unusable indexes. Otherwise, an error will be reported.
    • The target partition name must either be the same as the name of the last source partition or different from all partition names of the table.
    • The boundaries of the target partition are the union of the boundaries of all the source partitions.
    • For a range partitioned table, all source partitions must have contiguous boundaries.
    • For list partitioning, if the source partition contains a DEFAULT partition, the boundary of the target partition is also DEFAULT.
  • The syntax of modify_clause is used to set whether a partition index is usable.

    1	MODIFY PARTITION partition_name { UNUSABLE LOCAL INDEXES | REBUILD UNUSABLE LOCAL INDEXES }

  • The split_clause syntax is used to split one partition into partitions.

    The split_clause syntax for range partitioning is as follows:

    1	SPLIT PARTITION { partition_name | FOR ( partition_value [, ...] ) } { split_point_clause | no_split_point_clause }

    • The syntax of split_point_clause is as follows:

      1	AT ( partition_value ) INTO ( PARTITION partition_name , PARTITION partition_name )

      

      The size of split point should be in the range of splitting partition key. The split point can only split one partition into two.

    • The syntax of no_split_point_clause is as follows:

      1	INTO { ( partition_less_than_item [, ...] ) | ( partition_start_end_item [, ...] ) }

      

      • The first new partition key specified by partition_less_than_item must be larger than that of the former partition (if any), and the last partition key specified by partition_less_than_item must be equal to that of the splitting partition.
      • The start point (if any) of the first new partition specified by partition_start_end_item must be equal to the partition key (if any) of the previous partition. The end point (if any) of the last partition specified by partition_start_end_item must be equal to the partition key of the splitting partition.
      • partition_less_than_item supports a maximum of four partition keys and partition_start_end_item supports only one partition key. For details about the supported data types, see Partition Key.
      • partition_less_than_item and partition_start_end_item cannot be used in the same statement.
    • The syntax of partition_less_than_item is as follows:

      1 2	PARTITION partition_name VALUES LESS THAN ( { partition_value | MAXVALUE } [, ...] )

    • The syntax of partition_start_end_item is as follows. For details about the constraints, see partition_start_end_item syntax.

      1 2 3 4 5 6

      PARTITION partition_name { {START(partition_value) END (partition_value) EVERY (interval_value)} | {START(partition_value) END ({partition_value | MAXVALUE})} | {START(partition_value)} | {END({partition_value | MAXVALUE})} }

    The syntax of split_clause for list partitioning is as follows:

    1	SPLIT PARTITION { partition_name | FOR ( partition_value [, ...] ) } { split_values_clause | split_no_values_clause }

    • The syntax of split_values_clause that specifies a split point is as follows:

      1	VALUES ( { (partition_value) [, ...] } | DEFAULT } ) INTO ( PARTITION partition_name , PARTITION partition_name )

      

      • If the source partition is not a DEFAULT partition, the boundary specified by the split point is a non-void proper subset of the source partition boundary. If the source partition is a DEFAULT partition, the boundary specified by the split point cannot overlap with the boundaries of other non-DEFAULT partitions.
      • The boundary specified by the split point is the boundary of the first partition after the keyword INTO. The difference between the boundary of the source partition and the specified boundary of the split point is the boundary of the second partition.
      • If the source partition is the DEFAULT partition, the boundary of the second partition is still DEFAULT.
    • The syntax of split_no_values_clause that specifies no split points is as follows:

      1	INTO ( list_partition_item [, ....], PARTITION partition_name )

      

      • The syntax of list_partition_item is the same as the syntax specifying a partition in creating a list partitioned table, except that the boundary value here cannot be DEFAULT.
      • Except for the last partition, the boundaries of other partitions must be explicitly defined. The defined boundary cannot be DEFAULT and must be a non-empty proper subset of the source partition boundary. The boundary of the last partition is the difference set between the source partition boundary and other partition boundaries, and the boundary of the last partition is not empty (that is, the difference set cannot be empty).
      • If the source partition is a DEFAULT partition, the boundary of the last partition is DEFAULT.
  • The syntax of add_clause is used to add a partition to one or more specified partitioned tables.
    The add_clause syntax in range partitioning is as follows:

    1	ADD { partition_less_than_item... | partition_start_end_item }

    

    • The partition_less_than_item syntax can only be used for range partitioned tables. Otherwise, an error will be reported.
    • The syntax of partition_less_than_item is the same as the syntax specifying partitions in creating a range partitioned table.
    • If the boundary value of the last partition is a MAXVALUE, new partitions cannot be added. Otherwise, an error will be reported.

    The add_clause syntax for list partitioning is as follows:

    1	ADD list_partition_item

    

    • The list_partition_item syntax can only be used for a list partitioned table. Otherwise, an error will be reported.
    • The list_partition_item syntax is the same as the syntax specifying a partition in creating a list partitioned table.
    • If the current partition table contains DEFAULT partitions, no new partitions can be added. Otherwise, an error will be reported.
  • The syntax of drop_clause is used to remove a specified partition from a partitioned table.

    1	DROP PARTITION { partition_name | FOR ( partition_value [, ...] ) }

  • The drop_clause syntax supports deleting multiple partitions. (supported by clusters of 8.1.3.100 and later versions)

    1	DROP PARTITION { partition_name [, ... ] }

• The syntax of modifying a table partition name is as follows:

  1 2	ALTER TABLE [ IF EXISTS ] { table_name [*] | ONLY table_name | ONLY ( table_name )} RENAME PARTITION { partition_name | FOR ( partition_value [, ...] ) } TO partition_new_name;

Parameter Description

• table_name

  Specifies the name of a partitioned table.

  Value range: an existing partitioned table name

• partition_name

  Specifies the name of a partition.

  Value range: an existing partition name

• partition_value

  Specifies the key value of a partition.

  The value specified by PARTITION FOR ( partition_value [, ...] ) can uniquely identify a partition.

  Value range: value range of the partition key for the partition to be renamed

• UNUSABLE LOCAL INDEXES

  Sets all the indexes unusable in the partition.

• REBUILD UNUSABLE LOCAL INDEXES

  Rebuilds all the indexes in the partition.

• ENABLE/DISABLE ROW MOVEMENT

  Specifies the row movement switch.

  If the tuple value is updated on the partition key during the UPDATE action, the partition where the tuple is located is altered. Setting of this parameter enables error messages to be reported or movement of the tuple between partitions.

  Valid value:

  • ENABLE: The row movement switch is enabled.
  • DISABLE: The row movement switch is disabled.

  The switch is disabled by default.

• ordinary_table_name

  Specifies the name of the ordinary table whose data is to be migrated.

  Value range: an existing ordinary table name

• { WITH | WITHOUT } VALIDATION

  Checks whether the ordinary table data meets the specified partition key range of the partition to be migrated.

  Valid value:

  • WITH: checks whether the common table data meets the partition key range of the partition to be exchanged. If any data does not meet the required range, an error is reported.
  • WITHOUT: does not check whether the common table data meets the partition key range of the partition to be exchanged.

  The default value is WITH.

  The check is time consuming, especially when the data volume is large. Therefore, use WITHOUT when you are sure that the current common table data meets the partition key range of the partition to be exchanged.

• VERBOSE

  When VALIDATION is WITH, if the ordinary table contains data that is out of the partition key range, insert the data to the correct partition. If there is no correct partition where the data can be route to, an error is reported.

  

  Only when VALIDATION is WITH, VERBOSE can be specified.

• partition_new_name

  Specifies the new name of a partition.

  Value range: a string. It must comply with the naming convention.

Example

Delete partition P8:

1

ALTER TABLE tpcds.web_returns_p1 DROP PARTITION P8;

Delete partitions P3, P4, and P5:

1

ALTER TABLE tpcds.web_returns_p1 DROP PARTITION P3, P4, P5;

Add a partition WR_RETURNED_DATE_SK within the range 2453005 to 2453105:

1

ALTER TABLE tpcds.web_returns_p1 ADD PARTITION P8 VALUES LESS THAN (2453105);

Add a partition WR_RETURNED_DATE_SK within the range 2453105 to MAXVALUE:

1

ALTER TABLE tpcds.web_returns_p1 ADD PARTITION P9 VALUES LESS THAN (MAXVALUE);

Delete partition P8:

1

ALTER TABLE tpcds.web_returns_p1 DROP PARTITION FOR (2453005);

Rename the P7 partition as P10:

1

ALTER TABLE tpcds.web_returns_p1 RENAME PARTITION P7 TO P10;

Rename the P6 partition as P11:

1

ALTER TABLE tpcds.web_returns_p1 RENAME PARTITION FOR (2452639) TO P11;

Query rows in the P10 partition:

1
2
3
4
5

SELECT count(*) FROM tpcds.web_returns_p1 PARTITION (P10);
 count  
--------
 9362
(1 row)

Create a partitioned table using LIKE:

1

CREATE TABLE tpcds.web_returns_p3 (LIKE tpcds.web_returns_p2 INCLUDING PARTITION);

Split the P8 partition at 2453010:

1
2
3
4
5

ALTER TABLE tpcds.web_returns_p2 SPLIT PARTITION P8 AT (2453010) INTO
(
        PARTITION P9,
        PARTITION P10
); 

Merge the P6 and P7 partitions into one:

1

ALTER TABLE tpcds.web_returns_p2 MERGE PARTITIONS P6, P7 INTO PARTITION P8;

Modify the migration attribute of a partitioned table:

1

ALTER TABLE tpcds.web_returns_p2 DISABLE ROW MOVEMENT;

Add partitions [5000, 5300), [5300, 5600), [5600, 5900), and [5900, 6000):

1

ALTER TABLE tpcds.startend_pt ADD PARTITION p6 START(5000) END(6000) EVERY(300) TABLESPACE startend_tbs4;

Add the partition p7, specified by MAXVALUE:

1

ALTER TABLE tpcds.startend_pt ADD PARTITION p7 END(MAXVALUE);

Split the partition [4000, 5000) where 4500 is located:

1

ALTER TABLE tpcds.startend_pt SPLIT PARTITION FOR(4500) INTO(PARTITION q1 START(4000) END(5000) EVERY(250) TABLESPACE startend_tbs3);

Helpful Links

CREATE TABLE PARTITION, DROP TABLE