一张表内的数据过多时,就会严重影响到数据的查询和操作效率。openGauss支持把一张表从逻辑上分成多个小的分片,从而避免一次处理大量数据,提高处理效率。
openGauss数据库支持这些划分类型:
分区表的操作除了创建之外还有:
范围分区表按照划分范围的方式,分为以下类别:
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
)
PARTITION BY RANGE (partition_key)
(
PARTITION partition_name VALUES LESS THAN (partition_value | MAXVALUE})
[, ... ]
);
partition_table_name
分区表的名称。
column_name
新表中要创建的字段名。
data_type
字段的数据类型。
partition_key
partition_key为分区键的名称。
对于从句是VALUE LESS THAN的语法格式,范围分区策略的分区键最多支持4列。
partition_name
partition_name为范围分区的名称。
VALUES LESS THAN
分区中的数值必须小于上边界值。
partition_value
partition_value为范围分区的上边界,取值依赖于partition_key的类型。
MAXVALUE
MAXVALUE表示分区的上边界,它通常用于设置最后一个范围分区的上边界。
示例1:创建范围分区表sales_table,含有4个分区,分区键为DATE类型。分区的范围分别为:sales_date<2021-04-01,2021-04-01<= sales_date<2021-07-01,2021-07-01<=sales_date< 2021-10-01,2021-10-01 <= sales_date< MAXVALUE。
--创建分区表sales_table。
openGauss=# CREATE TABLE sales_table
(
order_no INTEGER NOT NULL,
goods_name CHAR(20) NOT NULL,
sales_date DATE NOT NULL,
sales_volume INTEGER,
sales_store CHAR(20)
)
PARTITION BY RANGE(sales_date)
(
PARTITION season1 VALUES LESS THAN('2021-04-01 00:00:00'),
PARTITION season2 VALUES LESS THAN('2021-07-01 00:00:00'),
PARTITION season3 VALUES LESS THAN('2021-10-01 00:00:00'),
PARTITION season4 VALUES LESS THAN(MAXVALUE)
);
-- 数据插入分区season1
openGauss=# INSERT INTO sales_table VALUES(1, 'jacket', '2021-01-10 00:00:00', 3,'Alaska');
-- 数据插入分区season2
openGauss=# INSERT INTO sales_table VALUES(2, 'hat', '2021-05-06 00:00:00', 5,'Clolorado');
-- 数据插入分区season3
openGauss=# INSERT INTO sales_table VALUES(3, 'shirt', '2021-09-17 00:00:00', 7,'Florida');
-- 数据插入分区season4
openGauss=# INSERT INTO sales_table VALUES(4, 'coat', '2021-10-21 00:00:00', 9,'Hawaii');
SELECT * FROM partition_table_name PARTITION { ( partition_name ) | FOR ( partition_value [, ...] ) }
partition_table_name
分区表的名称。
partition_name
partition_name为分区的名称。
partition_value
用于指定分区的值。PARTITION FOR子句指定的值所在的分区,就是进行查询的分区。
示例2:查询示例1中建立的分区表sales_table。
--查询sales_table的数据。
openGauss=# SELECT * FROM sales_table;
order_no | goods_name | sales_date | sales_volume | sale
s_store
----------+----------------------+---------------------+--------------+---------
-------------
1 | jacket | 2021-01-10 00:00:00 | 3 | Alaska
2 | hat | 2021-05-06 00:00:00 | 5 | Clolorado
3 | shirt | 2021-09-17 00:00:00 | 7 | Florida
4 | coat | 2021-10-21 00:00:00 | 9 | Hawaii
(4 rows)
--查询sales_table的4季度数据。这里采用“sales_table PARTITION (season4);”来引用第4季度数据所在分区。
openGauss=# SELECT * FROM sales_table PARTITION (season4);
order_no | goods_name | sales_date | sales_volume | sales_store
----------+----------------------+---------------------+--------------+----------------------
4 | coat | 2021-10-21 00:00:00 | 9 | Hawaii
(1 row)
--查询sales_table的1季度数据。这里采用“sales_table PARTITION FOR ('2021-3-21 00:00:00')”来引用第1季度数据所在分区。其中的'2021-3-21 00:00:00'处于第1季度所在分区。
openGauss=# SELECT * FROM sales_table PARTITION FOR ('2021-3-21 00:00:00');
order_no | goods_name | sales_date | sales_volume | sales_store
----------+----------------------+---------------------+--------------+----------------------
1 | jacket | 2021-01-10 00:00:00 | 3 | Alaska
(1 row)
START END范围分区表有多种表达方式,而且这些方式可以在一个分区表内组合使用。
方式一:START(partition_value) END (partition_value | MAXVALUE)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
)
PARTITION BY RANGE (partition_key)
(
PARTITION partition_name START(partition_value) END (partition_value | MAXVALUE)
[, ... ]
);
方式二:START(partition_value)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
] )
PARTITION BY RANGE (partition_key)
(
PARTITION partition_name START(partition_value)
[, ... ]
);
方式三:END(partition_value | MAXVALUE)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
] )
PARTITION BY RANGE (partition_key)
(
PARTITION partition_name END(partition_value | MAXVALUE)
[, ... ]
);
方式四:START(partition_value) END (partition_value) EVERY (interval_value)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
] )
PARTITION BY RANGE (partition_key)
(
PARTITION partition_name START(partition_value) END (partition_value) EVERY (interval_value)
[, ... ]
);
partition_table_name
分区表的名称。
column_name
新表中要创建的字段名。
data_type
字段的数据类型。
partition_key
partition_key为分区键的名称。
对于从句是START END的语法格式,范围分区策略的分区键仅支持1列。
partition_name
partition_name为范围分区的名称或者范围分区的名称前缀。
若该定义是“START(partition_value) END (partition_value) EVERY (interval_value)”从句,假定其中的partition_name是p1,则分区的名称依次为p1_1, p1_2, …。
例如对于定义“PARTITION p1 START(1) END(4) EVERY(1)”,则生成的分区是:[1, 2), [2, 3) 和 [3, 4),名称依次为p1_1, p1_2和p1_3,即此处的p1是名称前缀。
若该定义是第一个分区定义,且该定义有START值,则范围(MINVALUE, START)将自动作为第一个实际分区,其名称为p1_0,然后该定义语义描述的分区名称依次为p1_1, p1_2, …。
例如对于完整定义“PARTITION p1 START(1), PARTITION p2 START(2)”,生成的分区是:(MINVALUE, 1), [1, 2) 和 [2, MAXVALUE),其名称依次为p1_0, p1_1和p2,即此处p1是名称前缀,p2是分区名称。这里MINVALUE表示最小值。
其余的情况都是范围分区名称。
VALUES LESS THAN
分区中的数值必须小于上边界值。
partition_value
partition_value为范围分区的端点值(起始或终点),取值依赖于partition_key的类型。
interval_value:
对[START,END) 表示的范围进行切分,interval_value是指定切分后每个分区的宽度。如果(END-START)值不能整除以EVERY值,则仅最后一个分区的宽度小于EVERY值。
MAXVALUE
MAXVALUE表示分区的上边界,它通常用于设置最后一个范围分区的上边界。
示例3:以“START(partition_value) END (partition_value | MAXVALUE)”方式创建START END范围分区表graderecord。含有3个分区,分区键为INTEGER类型。分区的范围分别为:0<= grade<60,60<= grade<90,90<=grade< MAXVALUE。
--创建分区表graderecord。
openGauss=# CREATE TABLE graderecord
(
number INTEGER,
name CHAR(20),
class CHAR(20),
grade INTEGER
)
PARTITION BY RANGE(grade)
(
PARTITION pass START(60) END(90),
PARTITION excellent START(90) END(MAXVALUE)
);
-- 数据插入分区。
openGauss=# insert into graderecord values('210101','Alan','21.01',92);
openGauss=# insert into graderecord values('210102','Ben','21.01',62);
openGauss=# insert into graderecord values('210103','Brain','21.01',26);
openGauss=# insert into graderecord values('210204','Carl','21.02',77);
openGauss=# insert into graderecord values('210205','David','21.02',47);
openGauss=# insert into graderecord values('210206','Eric','21.02',97);
openGauss=# insert into graderecord values('210307','Frank','21.03',90);
openGauss=# insert into graderecord values('210308','Gavin','21.03',100);
openGauss=# insert into graderecord values('210309','Henry','21.03',67);
openGauss=# insert into graderecord values('210410','Jack','21.04',75);
openGauss=# insert into graderecord values('210311','Jerry','21.04',60);
--查询graderecord的数据。
openGauss=# SELECT * FROM graderecord;
number | name | class | grade
--------+----------------------+----------------------+-------
210103 | Brain | 21.01 | 26
210205 | David | 21.02 | 47
210102 | Ben | 21.01 | 62
210204 | Carl | 21.02 | 77
210309 | Henry | 21.03 | 67
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
210101 | Alan | 21.01 | 92
210206 | Eric | 21.02 | 97
210307 | Frank | 21.03 | 90
210308 | Gavin | 21.03 | 100
(11 rows)
--查询graderecord的pass分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (pass);
ERROR: partition "pass" of relation "graderecord" does not exist
查询失败。
原因是“PARTITION pass START(60) END(90),”是第一个分区定义,且该定义有START值,则范围(MINVALUE, 60)将自动作为第一个实际分区,其名称为“pass_0”。
而该定义语义描述的“60<= grade<90”分区的名称为“pass_1”。
--查询graderecord的pass_0分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (pass_0);
number | name | class | grade
--------+----------------------+----------------------+-------
210103 | Brain | 21.01 | 26
210205 | David | 21.02 | 47
(2 rows)
--查询graderecord的pass_1分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (pass_1);
number | name | class | grade
--------+----------------------+----------------------+-------
210102 | Ben | 21.01 | 62
210204 | Carl | 21.02 | 77
210309 | Henry | 21.03 | 67
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
(5 rows)
--查询graderecord的execllent分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (excellent);
number | name | class | grade
--------+----------------------+----------------------+-------
210101 | Alan | 21.01 | 92
210206 | Eric | 21.02 | 97
210307 | Frank | 21.03 | 90
210308 | Gavin | 21.03 | 100
(4 rows)
示例4:以“START(partition_value) END (partition_value) EVERY (interval_value)”方式创建START END范围分区表metro_ride_record。含有7个分区,分区键为INTEGER类型。总范围是ride_stations_number<21, 每3站为一个分区。
--创建分区表metro_ride_record。记录乘车人、上下车站点、乘坐站点数量。并按照乘坐站点数量,每3站建立一个分区。
openGauss=# CREATE TABLE metro_ride_record
(
record_number INTEGER,
name CHAR(20),
enter_station CHAR(20),
leave_station CHAR(20),
ride_stations_number INTEGER
)
PARTITION BY RANGE(ride_stations_number)
(
PARTITION cost START(3) END(21) EVERY (3)
);
-- 数据插入分区。
openGauss=# insert into metro_ride_record values('120101','Brain','Tung Chung','Tsing Yi',2);
openGauss=# insert into metro_ride_record values('120102','David','Po Lam','Yau Tong',4);
openGauss=# insert into metro_ride_record values('120103','Ben','Yau Ma Tei','Wong Tai Sin',6);
openGauss=# insert into metro_ride_record values('120104','Carl','Tai Wo Hau','Prince Edward',8);
openGauss=# insert into metro_ride_record values('120105','Henry','Admiralty','Lai King',10);
openGauss=# insert into metro_ride_record values('120106','Jack','Chai Wan','Central',12);
openGauss=# insert into metro_ride_record values('120107','Jerry','Central','Tai Wo Hau',14);
openGauss=# insert into metro_ride_record values('120108','Alan','Diamond Hill','Kwai Hing',16);
openGauss=# insert into metro_ride_record values('120109','Eric','Jordan','Shek Kip Mei',18);
openGauss=# insert into metro_ride_record values('120110','Frank','Lok Fu','Sunny Bay',20);
--查询metro_ride_record的数据。
openGauss=# SELECT * FROM metro_ride_record;
record_number | name | enter_station | leave_station | ride_stations_number
---------------+----------------------+----------------------+----------------------+----------------------
120101 | Brain | Tung Chung | Tsing Yi | 2
120102 | David | Po Lam | Yau Tong | 4
120103 | Ben | Yau Ma Tei | Wong Tai Sin | 6
120104 | Carl | Tai Wo Hau | Prince Edward | 8
120105 | Henry | Admiralty | Lai King | 10
120106 | Jack | Chai Wan | Central | 12
120107 | Jerry | Central | Tai Wo Hau | 14
120108 | Alan | Diamond Hill | Kwai Hing | 16
120109 | Eric | Jordan | Shek Kip Mei | 18
120110 | Frank | Lok Fu | Sunny Bay | 20
(10 rows)
“PARTITION cost START(3) END(21) EVERY (3)”是第一个分区定义,且该定义有START值,则范围(MINVALUE, 3)将自动作为第一个实际分区,其名称为“cost_0”。
其余分区依次为“cost_1”、...、“cost_6”.
--查询metro_ride_record的cost_0分区数据。
openGauss=# SELECT * FROM metro_ride_record PARTITION (cost_0);
record_number | name | enter_station | leave_station | ride_stations_number
---------------+----------------------+----------------------+----------------------+----------------------
120101 | Brain | Tung Chung | Tsing Yi | 2
(1 row)
--查询metro_ride_record的cost_1分区数据。
openGauss=# SELECT * FROM metro_ride_record PARTITION (cost_1);
record_number | name | enter_station | leave_station | ride_stations_number
---------------+----------------------+----------------------+----------------------+----------------------
120102 | David | Po Lam | Yau Tong | 4
(1 row)
--查询metro_ride_record的cost_6分区数据。
openGauss=# SELECT * FROM metro_ride_record PARTITION (cost_6);
record_number | name | enter_station | leave_station | ride_stations_number
---------------+----------------------+----------------------+----------------------+----------------------
120109 | Eric | Jordan | Shek Kip Mei | 18
120110 | Frank | Lok Fu | Sunny Bay | 20
(2 rows)
示例5:以“START(partition_value) ”方式创建START END范围分区表graderecord。含有3个分区,分区键为INTEGER类型。分区的范围分别为:0<= grade<60,60<= grade<90,90<=grade< MAXVALUE。
--创建分区表graderecord。
openGauss=# CREATE TABLE graderecord
(
number INTEGER,
name CHAR(20),
class CHAR(20),
grade INTEGER
)
PARTITION BY RANGE(grade)
(
PARTITION pass START(60),
PARTITION excellent START(90)
);
-- 数据插入分区。
openGauss=# insert into graderecord values('210101','Alan','21.01',92);
openGauss=# insert into graderecord values('210102','Ben','21.01',62);
openGauss=# insert into graderecord values('210103','Brain','21.01',26);
openGauss=# insert into graderecord values('210204','Carl','21.02',77);
openGauss=# insert into graderecord values('210205','David','21.02',47);
openGauss=# insert into graderecord values('210206','Eric','21.02',97);
openGauss=# insert into graderecord values('210307','Frank','21.03',90);
openGauss=# insert into graderecord values('210308','Gavin','21.03',100);
openGauss=# insert into graderecord values('210309','Henry','21.03',67);
openGauss=# insert into graderecord values('210410','Jack','21.04',75);
openGauss=# insert into graderecord values('210311','Jerry','21.04',60);
--查询graderecord的数据。
openGauss=# SELECT * FROM graderecord;
number | name | class | grade
--------+----------------------+----------------------+-------
210103 | Brain | 21.01 | 26
210205 | David | 21.02 | 47
210102 | Ben | 21.01 | 62
210204 | Carl | 21.02 | 77
210309 | Henry | 21.03 | 67
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
210101 | Alan | 21.01 | 92
210206 | Eric | 21.02 | 97
210307 | Frank | 21.03 | 90
210308 | Gavin | 21.03 | 100
(11 rows)
--查询graderecord的pass分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (pass);
ERROR: partition "pass" of relation "graderecord" does not exist
查询失败。
原因是“PARTITION pass START(60),”是第一个分区定义,且该定义有START值,则范围(MINVALUE, 60)将自动作为第一个实际分区,其名称为“pass_0”。
而该定义语义描述的“60<= grade<90”分区的名称为“pass_1”。
--查询graderecord的pass_0分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (pass_0);
number | name | class | grade
--------+----------------------+----------------------+-------
210103 | Brain | 21.01 | 26
210205 | David | 21.02 | 47
(2 rows)
--查询graderecord的pass_1分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (pass_1);
number | name | class | grade
--------+----------------------+----------------------+-------
210102 | Ben | 21.01 | 62
210204 | Carl | 21.02 | 77
210309 | Henry | 21.03 | 67
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
(5 rows)
--查询graderecord的execllent分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (excellent);
number | name | class | grade
--------+----------------------+----------------------+-------
210101 | Alan | 21.01 | 92
210206 | Eric | 21.02 | 97
210307 | Frank | 21.03 | 90
210308 | Gavin | 21.03 | 100
(4 rows)
示例6:以“END(partition_value | MAXVALUE) ”方式创建START END范围分区表graderecord。含有3个分区,分区键为INTEGER类型。分区的范围分别为:0<= grade<60,60<= grade<90,90<=grade< MAXVALUE。。
--创建分区表graderecord。
openGauss=# CREATE TABLE graderecord
(
number INTEGER,
name CHAR(20),
class CHAR(20),
grade INTEGER
)
PARTITION BY RANGE(grade)
(
PARTITION no_pass END(60),
PARTITION pass END(90),
PARTITION excellent END(MAXVALUE)
);
-- 数据插入分区。
openGauss=# insert into graderecord values('210101','Alan','21.01',92);
openGauss=# insert into graderecord values('210102','Ben','21.01',62);
openGauss=# insert into graderecord values('210103','Brain','21.01',26);
openGauss=# insert into graderecord values('210204','Carl','21.02',77);
openGauss=# insert into graderecord values('210205','David','21.02',47);
openGauss=# insert into graderecord values('210206','Eric','21.02',97);
openGauss=# insert into graderecord values('210307','Frank','21.03',90);
openGauss=# insert into graderecord values('210308','Gavin','21.03',100);
openGauss=# insert into graderecord values('210309','Henry','21.03',67);
openGauss=# insert into graderecord values('210410','Jack','21.04',75);
openGauss=# insert into graderecord values('210311','Jerry','21.04',60);
--查询graderecord的数据。
openGauss=# SELECT * FROM graderecord;
number | name | class | grade
--------+----------------------+----------------------+-------
210103 | Brain | 21.01 | 26
210205 | David | 21.02 | 47
210102 | Ben | 21.01 | 62
210204 | Carl | 21.02 | 77
210309 | Henry | 21.03 | 67
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
210101 | Alan | 21.01 | 92
210206 | Eric | 21.02 | 97
210307 | Frank | 21.03 | 90
210308 | Gavin | 21.03 | 100
(11 rows)
--查询graderecord的no_pass分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (no_pass);
number | name | class | grade
--------+----------------------+----------------------+-------
210103 | Brain | 21.01 | 26
210205 | David | 21.02 | 47
(2 rows)
--查询graderecord的pass分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (pass);
number | name | class | grade
--------+----------------------+----------------------+-------
210102 | Ben | 21.01 | 62
210204 | Carl | 21.02 | 77
210309 | Henry | 21.03 | 67
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
(5 rows)
--查询graderecord的execllent分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (excellent);
number | name | class | grade
--------+----------------------+----------------------+-------
210101 | Alan | 21.01 | 92
210206 | Eric | 21.02 | 97
210307 | Frank | 21.03 | 90
210308 | Gavin | 21.03 | 100
(4 rows)
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
)
PARTITION BY LIST (partition_key)
(
PARTITION partition_name VALUES (list_values_clause)
[, ... ]
);
partition_table_name
分区表的名称。
column_name
新表中要创建的字段名。
data_type
字段的数据类型。
partition_key
partition_key为分区键的名称。
列表分区策略的分区键仅支持1列。
partition_name
partition_name为范围分区的名称。
list_values_clause
对应分区存在的一个或者多个键值。多个键值之间以逗号分隔。
VALUES (DEFAULT)
加入的数据如有“list_values_clause”中未列出的键值,存放在VALUES (DEFAULT)对应的分区。
MAXVALUE
MAXVALUE表示分区的上边界,它通常用于设置最后一个范围分区的上边界。
示例7:创建列表分区表graderecord。含有4个分区,分区键为CHAR类型。分区的范围分别为:21.01,21.02,21.03,21.04。
--创建分区表graderecord。
openGauss=# CREATE TABLE graderecord
(
number INTEGER,
name CHAR(20),
class CHAR(20),
grade INTEGER
)
PARTITION BY LIST(class)
(
PARTITION class_01 VALUES ('21.01'),
PARTITION class_02 VALUES ('21.02'),
PARTITION class_03 VALUES ('21.03'),
PARTITION class_04 VALUES ('21.04')
);
-- 数据插入分区。
openGauss=# insert into graderecord values('210101','Alan','21.01',92);
openGauss=# insert into graderecord values('210102','Ben','21.01',62);
openGauss=# insert into graderecord values('210103','Brain','21.01',26);
openGauss=# insert into graderecord values('210204','Carl','21.02',77);
openGauss=# insert into graderecord values('210205','David','21.02',47);
openGauss=# insert into graderecord values('210206','Eric','21.02',97);
openGauss=# insert into graderecord values('210307','Frank','21.03',90);
openGauss=# insert into graderecord values('210308','Gavin','21.03',100);
openGauss=# insert into graderecord values('210309','Henry','21.03',67);
openGauss=# insert into graderecord values('210410','Jack','21.04',75);
openGauss=# insert into graderecord values('210311','Jerry','21.04',60);
--查询graderecord的数据。
openGauss=# SELECT * FROM graderecord;
number | name | class | grade
--------+----------------------+----------------------+-------
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
210307 | Frank | 21.03 | 90
210308 | Gavin | 21.03 | 100
210309 | Henry | 21.03 | 67
210204 | Carl | 21.02 | 77
210205 | David | 21.02 | 47
210206 | Eric | 21.02 | 97
210101 | Alan | 21.01 | 92
210102 | Ben | 21.01 | 62
210103 | Brain | 21.01 | 26
(11 rows)
--查询graderecord的class_01分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (class_01);
number | name | class | grade
--------+----------------------+----------------------+-------
210101 | Alan | 21.01 | 92
210102 | Ben | 21.01 | 62
210103 | Brain | 21.01 | 26
(3 rows)
--查询graderecord的class_04分区数据。
openGauss=# SELECT * FROM graderecord PARTITION (class_04);
number | name | class | grade
--------+----------------------+----------------------+-------
210410 | Jack | 21.04 | 75
210311 | Jerry | 21.04 | 60
(2 rows)
间隔分区是在范围分区的基础上,增加了间隔值“PARTITION BY RANGE (partition_key)”的定义。
VALUES LESS THAN间隔分区语法格式:
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
)
PARTITION BY RANGE (partition_key)
(
INTERVAL ('interval_expr')
PARTITION partition_name VALUES LESS THAN (partition_value | MAXVALUE})
[, ... ]
);
START END间隔分区表语法格式:
方式一:START(partition_value) END (partition_value | MAXVALUE)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
)
PARTITION BY RANGE (partition_key)
(
INTERVAL ('interval_expr')
PARTITION partition_name START(partition_value) END (partition_value | MAXVALUE)
[, ... ]
);
方式二:START(partition_value) END (partition_value) EVERY (interval_value)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
] )
PARTITION BY RANGE (partition_key)
(
PARTITION partition_name START(partition_value) END (partition_value) EVERY (interval_value)
[, ... ]
);
方式三:START(partition_value)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
] )
PARTITION BY RANGE (partition_key)
(
INTERVAL ('interval_expr')
PARTITION partition_name START(partition_value)
[, ... ]
);
方式四:END(partition_value | MAXVALUE)方式
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
] )
PARTITION BY RANGE (partition_key)
INTERVAL ('interval_expr')
(
PARTITION partition_name END(partition_value | MAXVALUE)
[, ... ]
);
INTERVAL (‘interval_expr’)
间隔分区定义信息。只支持TIMESTAMP[§] [WITHOUT TIME ZONE]、TIMESTAMP[§] [WITH TIME ZONE]、DATE数据类型。
interval_expr自动创建分区的间隔,例如:
自动创建分区的间隔,例如:1 day、1 month。
partition_name
partition_name为范围分区的名称。
系统自动建立的分区按照建立的先后顺序,依次命名为:sys_p1、sys_p2、sys_p3…
示例8:间隔分区表sales_table。
--创建分区表sales_table。
openGauss=# CREATE TABLE sales_table
(
order_no INTEGER NOT NULL,
goods_name CHAR(20) NOT NULL,
sales_date DATE NOT NULL,
sales_volume INTEGER,
sales_store CHAR(20)
)
PARTITION BY RANGE(sales_date)
INTERVAL ('1 month')
(
PARTITION start VALUES LESS THAN('2021-01-01 00:00:00'),
PARTITION later VALUES LESS THAN('2021-01-10 00:00:00')
);
-- 数据插入分区later
openGauss=# INSERT INTO sales_table VALUES(1, 'jacket', '2021-01-8 00:00:00', 3,'Alaska');
-- 不在已有分区的数据插入,系统会新建分区sys_p1。
openGauss=# INSERT INTO sales_table VALUES(2, 'hat', '2021-04-06 00:00:00', 255,'Clolorado');
-- 不在已有分区的数据插入,系统会新建分区sys_p2。
openGauss=# INSERT INTO sales_table VALUES(3, 'shirt', '2021-11-17 00:00:00', 7000,'Florida');
-- 数据插入分区start
openGauss=# INSERT INTO sales_table VALUES(4, 'coat', '2020-10-21 00:00:00', 9000,'Hawaii');
--查询sales_table的数据。
openGauss=# SELECT * FROM sales_table;
order_no | goods_name | sales_date | sales_volume | sales_store
----------+----------------------+---------------------+--------------+----------------------
4 | coat | 2020-10-21 00:00:00 | 9000 | Hawaii
1 | jacket | 2021-01-08 00:00:00 | 3 | Alaska
2 | hat | 2021-04-06 00:00:00 | 255 | Clolorado
3 | shirt | 2021-11-17 00:00:00 | 7000 | Florida
(4 rows)
--查询sales_table的start分区数据。这里采用“sales_table PARTITION (start);”来引用分区。
openGauss=# SELECT * FROM sales_table PARTITION (start);
order_no | goods_name | sales_date | sales_volume | sales_store
----------+----------------------+---------------------+--------------+----------------------
4 | coat | 2020-10-21 00:00:00 | 9000 | Hawaii
(1 row)
--查询sales_table的later分区数据。这里采用“sales_table PARTITION (later);”来引用分区。
openGauss=# SELECT * FROM sales_table PARTITION (later);
order_no | goods_name | sales_date | sales_volume | sales_store
----------+----------------------+---------------------+--------------+----------------------
1 | jacket | 2021-01-08 00:00:00 | 3 | Alaska
(1 row)
--查询sales_table的sys_p1分区数据。这里采用“sales_table PARTITION (sys_p1);”来引用分区。
openGauss=# SELECT * FROM sales_table PARTITION (sys_p1);
order_no | goods_name | sales_date | sales_volume | sales_store
----------+----------------------+---------------------+--------------+----------------------
2 | hat | 2021-04-06 00:00:00 | 255 | Clolorado
(1 row)
--查询sales_table的sys_p2分区数据。这里采用“sales_table PARTITION (sys_p2);”来引用分区。
openGauss=# SELECT * FROM sales_table PARTITION (sys_p2);
order_no | goods_name | sales_date | sales_volume | sales_store
----------+----------------------+---------------------+--------------+----------------------
3 | shirt | 2021-11-17 00:00:00 | 7000 | Florida
(1 row)
CREATE TABLE partition_table_name
( [column_name data_type ]
[, ... ]
)
PARTITION BY HASH (partition_key)
(PARTITION partition_name )
[, ... ]
);
partition_table_name
分区表的名称。
column_name
新表中要创建的字段名。
data_type
字段的数据类型。
partition_key
partition_key为分区键的名称。哈希分区策略的分区键仅支持1列。
partition_name
partition_name为哈希分区的名称。希望创建几个哈希分区就给出几个分区名。
示例9:哈希分区表hash_partition_table。
--创建哈希分区表hash_partition_table
openGauss=# create table hash_partition_table (
col1 int,
col2 int)
partition by hash(col1)
(
partition p1,
partition p2
);
-- 数据插入
openGauss=# INSERT INTO hash_partition_table VALUES(1, 1);
INSERT 0 1
openGauss=# INSERT INTO hash_partition_table VALUES(2, 2);
INSERT 0 1
openGauss=# INSERT INTO hash_partition_table VALUES(3, 3);
INSERT 0 1
openGauss=# INSERT INTO hash_partition_table VALUES(4, 4);
INSERT 0 1
-- 查看数据
openGauss=# select * from hash_partition_table partition (p1);
col1 | col2
------+------
3 | 3
4 | 4
(2 rows)
openGauss=# select * from hash_partition_table partition (p2);
col1 | col2
------+------
1 | 1
2 | 2
(2 rows)
导入单行数据:
INSERT INTO partition_table_name [ ( column_name [, ...] ) ] VALUES [ ( value )[, ...] ];
导入结构相同的现有表格数据:
INSERT INTO partition_table_name SELECT * FROM source_table_name
partition_table_name
分区表的名称。
column_name
分区表中的字段名。可省略。
value
字段对应的值:
示例10:
--创建分区表employees_table。
openGauss=# CREATE TABLE employees_table
(
employee_id INTEGER NOT NULL,
employee_name CHAR(20) NOT NULL,
onboarding_date DATE NOT NULL,
position CHAR(20)
)
PARTITION BY RANGE(onboarding_date)
(
PARTITION founders VALUES LESS THAN('2000-01-01 00:00:00'),
PARTITION senate VALUES LESS THAN('2010-01-01 00:00:00'),
PARTITION seniors VALUES LESS THAN('2020-01-01 00:00:00'),
PARTITION newcomer VALUES LESS THAN(MAXVALUE)
);
-- 数据插入分区founders
openGauss=# INSERT INTO employees_table VALUES(1, 'SMITH', '1997-01-10 00:00:00','Manager');
-- 查看founders分区数据
openGauss=# select * from employees_table partition (founders);
-- 创建表格employees_data_table
openGauss=# CREATE TABLE employees_data_table
(
employee_id INTEGER NOT NULL,
employee_name CHAR(20) NOT NULL,
onboarding_date DATE NOT NULL,
position CHAR(20)
);
-- 插入数据
openGauss=# insert into employees_data_table (employee_id, employee_name, onboarding_date, position) VALUES
(2, 'JONES', '2001-05-06 00:00:00', 'Supervisor'),
(3, 'WILLIAMS', '2011-09-17 00:00:00', 'Engineer'),
(4, 'TAYLOR', '2021-10-21 00:00:00', 'Clerk');
-- 查看表格数据
openGauss=# select * from employees_data_table;
--数据导入employees_table
openGauss=# INSERT INTO employees_table SELECT * FROM employees_data_table;
-- 查看senate分区数据
openGauss=# select * from employees_table partition (senate);
employee_id | employee_name | onboarding_date | position
-------------+----------------------+---------------------+----------------------
2 | JONES | 2001-05-06 00:00:00 | Supervisor
(1 row)
-- 查看seniors分区数据
openGauss=# select * from employees_table partition (seniors);
employee_id | employee_name | onboarding_date | position
-------------+----------------------+---------------------+----------------------
3 | WILLIAMS | 2011-09-17 00:00:00 | Engineer
(1 row)
-- 查看newcomer分区数据
openGauss=# select * from employees_table partition (newcomer);
employee_id | employee_name | onboarding_date | position
-------------+----------------------+---------------------+----------------------
4 | TAYLOR | 2021-10-21 00:00:00 | Clerk
(1 row)
删除分区:
ALTER TABLE partition_table_name DROP PARTITION partition_name;
增加分区:
ALTER TABLE partition_table_name ADD {partition_less_than_item | partition_start_end_item| partition_list_item };
重命名分区:
ALTER TABLE partition_table_name RENAME PARTITION partition_name TO partition_new_name;
分裂分区(指定切割点split_partition_value的语法):
ALTER TABLE partition_table_name SPLIT PARTITION partition_name AT ( split_partition_value ) INTO ( PARTITION partition_new_name1, PARTITION partition_new_name2);
分裂分区(指定分区范围的语法):
ALTER TABLE partition_table_name SPLIT PARTITION partition_name INTO { ( partition_less_than_item [, ...] ) | ( partition_start_end_item [, ...] ) };
合并分区:
ALTER TABLE partition_table_name MERGE PARTITIONS { partition_name } [, ...] INTO PARTITION partition_name;
partition_table_name
分区表的名称。
partition_name
partition_name为分区的名称。
split_partition_value
切割点。
PARTITION partition_new_name1, PARTITION partition_new_name2
按照切割点分裂出的两个分区。
partition_less_than_item
分区项的描述语句,语法为:
PARTITION partition_name VALUES LESS THAN ( { partition_value | MAXVALUE } [, ...] )
用法与创建VALUES LESS THAN范围分区表语法格式中相同。
partition_start_end_item
分区项的描述语句,语法为:
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})}
用法与创建START END范围分区表语法格式中相同。
partition_list_item
分区项的描述语句,语法为:
PARTITION partition_name VALUES (list_values_clause)
用法与创建列表分区表语法格式中相同。
split_point_clause
分裂分区时,指定的切割点。
partition_value
分区键值。
示例11:
--创建分区表employees_table。
openGauss=# CREATE TABLE employees_table
(
employee_id INTEGER NOT NULL,
employee_name CHAR(20) NOT NULL,
onboarding_date DATE NOT NULL,
position CHAR(20)
)
PARTITION BY RANGE(onboarding_date)
(
PARTITION founders VALUES LESS THAN('2000-01-01 00:00:00'),
PARTITION senate VALUES LESS THAN('2010-01-01 00:00:00'),
PARTITION seniors VALUES LESS THAN('2020-01-01 00:00:00'),
PARTITION newcomer VALUES LESS THAN(MAXVALUE)
);
-- 插入数据
openGauss=# INSERT INTO employees_table VALUES
(1, 'SMITH', '1997-01-10 00:00:00','Manager'),
(2, 'JONES', '2001-05-06 00:00:00', 'Supervisor'),
(3, 'WILLIAMS', '2011-09-17 00:00:00', 'Engineer'),
(4, 'TAYLOR', '2021-10-21 00:00:00', 'Clerk');
查看newcomer分区
openGauss=# SELECT * FROM employees_table PARTITION (newcomer);
employee_id | employee_name | onboarding_date | position
-------------+----------------------+---------------------+----------------------
4 | TAYLOR | 2021-10-21 00:00:00 | Clerk
(1 row)
--删除newcomer分区。
openGauss=# ALTER TABLE employees_table DROP PARTITION newcomer;
ALTER TABLE
-- 查看newcomer分区数据
openGauss=# select * from employees_table partition (newcomer);
ERROR: partition "newcomer" of relation "employees_table" does not exist
--增加fresh分区。
openGauss=# ALTER TABLE employees_table ADD PARTITION fresh VALUES LESS THAN ('2040-01-01 00:00:00');
ALTER TABLE
--以2030-01-01 00:00:00为分割点,分裂fresh分区为current、future两个分区
openGauss=# ALTER TABLE employees_table SPLIT PARTITION fresh AT ('2030-01-01 00:00:00') INTO (PARTITION current, PARTITION future);
ALTER TABLE
--将分区current改名为now
openGauss=# ALTER TABLE employees_table RENAME PARTITION current TO now;
ALTER TABLE
--将founders,senate合并为一个分区original。
openGauss=# ALTER TABLE employees_table MERGE PARTITIONS founders, senate INTO PARTITION original;
DROP TABLE partition_table_name;
partition_table_name
分区表的名称。
示例12:
--删除分区表employees_table。
openGauss=# DROP TABLE employees_table;
DROP TABLE
点赞,你的认可是我创作的动力!
⭐️ 收藏,你的青睐是我努力的方向!
✏️ 评论,你的意见是我进步的财富!