PostgreSQL和PPAS的分区表及多种条件下的性能体现二

zhouer

　　PPAS中的分区表可以按oracle兼容的语法创建，具体使用请参见《Postgres_Plus_Advanced_Server_Oracle_Compatibility_Guide_v91.pdf》。
下面是分区表上操作的相关情况

1
创建表:
create table test (id integer primary key, name varchar(32))
PARTITION BY RANGE (id)
(PARTITION t1_1000 VALUES LESS THAN(1001),
 PARTITION t1001_2000 VALUES LESS THAN(2001),
 PARTITION t2001_3000 VALUES LESS THAN(3001));

1.1
从数据库取的表定义
-- Table: test
-- DROP TABLE test;
CREATE TABLE test
(
  id integer NOT NULL,
  name character varying(32),
  CONSTRAINT test_pkey PRIMARY KEY (id)
)
WITH (
  OIDS=FALSE
);
ALTER TABLE test
  OWNER TO enterprisedb;


-- Table: test_t1_1000
-- DROP TABLE test_t1_1000;
CREATE TABLE test_t1_1000
(
-- Inherited from table test:  id integer NOT NULL,
-- Inherited from table test:  name character varying(32),
  CONSTRAINT test_t1_1000_pkey PRIMARY KEY (id),
  CONSTRAINT test_t1_1000_partition CHECK (id < 1001)
)
INHERITS (test)
WITH (
  OIDS=FALSE
);
ALTER TABLE test_t1_1000
  OWNER TO enterprisedb;

-- Table: test_t1001_2000
-- DROP TABLE test_t1001_2000;
CREATE TABLE test_t1001_2000
(
-- Inherited from table test:  id integer NOT NULL,
-- Inherited from table test:  name character varying(32),
  CONSTRAINT test_t1001_2000_pkey PRIMARY KEY (id),
  CONSTRAINT test_t1001_2000_partition CHECK (id >= 1001 AND id < 2001)
)
INHERITS (test)
WITH (
  OIDS=FALSE
);
ALTER TABLE test_t1001_2000
  OWNER TO enterprisedb;


-- Table: test_t2001_3000
-- DROP TABLE test_t2001_3000;
CREATE TABLE test_t2001_3000
(
-- Inherited from table test:  id integer NOT NULL,
-- Inherited from table test:  name character varying(32),
  CONSTRAINT test_t2001_3000_pkey PRIMARY KEY (id),
  CONSTRAINT test_t2001_3000_partition CHECK (id >= 2001 AND id < 3001)
)
INHERITS (test)
WITH (
  OIDS=FALSE
);
ALTER TABLE test_t2001_3000
  OWNER TO enterprisedb;

2
给表中插入值时自动根据ID值插入到分区表中
edbtest=# INSERT INTO test(id, name)VALUES (6, 'ertr');
INSERT 0 0

edbtest=# select * from test;
 id | name
----+------
  6 | ertr
(1 row)

edbtest=#
edbtest=# select count(*) from only test;
 count
-------
     0
(1 row)

edbtest=# select count(*) from only test_t1_1000;
 count
-------
     1
(1 row)
                                  ^
edbtest=#
edbtest=# select count(*) from only test_t1001_2000;
 count
-------
     0
(1 row)

edbtest=#

3
从父表中删除该值
edbtest=# delete from test where id=6;
DELETE 1
edbtest=#
edbtest=# select count(*) from only test_t1_1000;
 count
-------
     0
(1 row)

4
批量插入值
edbtest=#insert into test select generate_series(1,2600),'abc';
INSERT 0 0
edbtest=#
edbtest=# select count(*) from test;
 count
-------
  2600
(1 row)

edbtest=# select count(*) from only test;
 count
-------
     0
(1 row)

edbtest=# select count(*) from only test_t2001_3000;
 count
-------
   600
(1 row)

edbtest=#

5
查询

5.1
按分区列值查询,只查询对应分区表
edbtest=# explain select * from test where id=200;
                                              QUERY PLAN                                              
-------------------------------------------------------------------------------------------------------
 Result  (cost=0.00..7.27 rows=2 width=47)
   ->  Append  (cost=0.00..7.27 rows=2 width=47)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: (id = 200)
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = 200)
(6 rows)

5.2
按分区列值做范围查询,只查询对应分区表
edbtest=# explain select * from test where id<200 and id>100;
                                               QUERY PLAN                                               
---------------------------------------------------------------------------------------------------------
 Result  (cost=0.00..9.25 rows=101 width=9)
   ->  Append  (cost=0.00..9.25 rows=101 width=9)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: ((id < 200) AND (id > 100))
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..9.25 rows=100 width=8)
               Index Cond: ((id < 200) AND (id > 100))
(6 rows)

edbtest=#

edbtest=# explain select * from test where id<700 and id>100;
                                   QUERY PLAN                                  
--------------------------------------------------------------------------------
 Result  (cost=0.00..20.00 rows=601 width=8)
   ->  Append  (cost=0.00..20.00 rows=601 width=8)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: ((id < 700) AND (id > 100))
         ->  Seq Scan on test_t1_1000 test  (cost=0.00..20.00 rows=600 width=8)
               Filter: ((id < 700) AND (id > 100))
(6 rows)

edbtest=#
edbtest=# explain select * from test where id<1100 and id>900;
                                                  QUERY PLAN                                                  
---------------------------------------------------------------------------------------------------------------
 Result  (cost=0.00..18.50 rows=201 width=8)
   ->  Append  (cost=0.00..18.50 rows=201 width=8)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: ((id < 1100) AND (id > 900))
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..9.25 rows=100 width=8)
               Index Cond: ((id < 1100) AND (id > 900))
         ->  Index Scan using test_t1001_2000_pkey on test_t1001_2000 test  (cost=0.00..9.25 rows=100 width=8)
               Index Cond: ((id < 1100) AND (id > 900))
(8 rows)

5.3
按分区列值和其它列查询,只查询对应分区表
edbtest=# explain select * from test where id=300 and name='ccc';
                                              QUERY PLAN                                              
-------------------------------------------------------------------------------------------------------
 Result  (cost=0.00..7.27 rows=2 width=47)
   ->  Append  (cost=0.00..7.27 rows=2 width=47)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: ((id = 300) AND ((name)::text = 'ccc'::text))
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = 300)
               Filter: ((name)::text = 'ccc'::text)
(7 rows)

5.4
按分区列值查询,值有显式类型转换,只查询对应分区表
edbtest=# explain select * from test where id='5'::int;
                                              QUERY PLAN                                              
-------------------------------------------------------------------------------------------------------
 Result  (cost=0.00..7.27 rows=2 width=47)
   ->  Append  (cost=0.00..7.27 rows=2 width=47)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: (id = 5)
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = 5)
(6 rows)

5.5
按分区列值查询,值和列类型不同,值有隐式类型转换,只查询对应分区表
edbtest=# explain select * from test where id='5';
                                              QUERY PLAN                                              
-------------------------------------------------------------------------------------------------------
 Result  (cost=0.00..7.27 rows=2 width=47)
   ->  Append  (cost=0.00..7.27 rows=2 width=47)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: (id = 5)
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = 5)
(6 rows)


5.6
按分区列值查询,列要做隐式类型转换,走全表扫描
edbtest=# explain select * from test where id || name ='5abc';
                                   QUERY PLAN                                   
---------------------------------------------------------------------------------
 Result  (cost=0.00..65.00 rows=14 width=14)
   ->  Append  (cost=0.00..65.00 rows=14 width=14)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: (((id)::text || (name)::text) = '5abc'::text)
         ->  Seq Scan on test_t1_1000 test  (cost=0.00..25.00 rows=5 width=8)
               Filter: (((id)::text || (name)::text) = '5abc'::text)
         ->  Seq Scan on test_t1001_2000 test  (cost=0.00..25.00 rows=5 width=8)
               Filter: (((id)::text || (name)::text) = '5abc'::text)
         ->  Seq Scan on test_t2001_3000 test  (cost=0.00..15.00 rows=3 width=8)
               Filter: (((id)::text || (name)::text) = '5abc'::text)
(10 rows)

edbtest=#

5.7
按分区列值查询,值使用了函数,走全表扫描
edbtest=# explain select * from test where id=to_number('1');
                                   QUERY PLAN                                   
---------------------------------------------------------------------------------
 Result  (cost=0.00..52.00 rows=14 width=14)
   ->  Append  (cost=0.00..52.00 rows=14 width=14)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: ((id)::numeric = 1::numeric)
         ->  Seq Scan on test_t1_1000 test  (cost=0.00..20.00 rows=5 width=8)
               Filter: ((id)::numeric = 1::numeric)
         ->  Seq Scan on test_t1001_2000 test  (cost=0.00..20.00 rows=5 width=8)
               Filter: ((id)::numeric = 1::numeric)
         ->  Seq Scan on test_t2001_3000 test  (cost=0.00..12.00 rows=3 width=8)
               Filter: ((id)::numeric = 1::numeric)
(10 rows)

5.8
按分区列值查询,值使用了函数,走分区表索引扫描
edbtest=# explain select * from test where id=cast('1' as int);
                                              QUERY PLAN                                              
-------------------------------------------------------------------------------------------------------
 Result  (cost=0.00..7.27 rows=2 width=47)
   ->  Append  (cost=0.00..7.27 rows=2 width=47)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: (id = 1)
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = 1)
(6 rows)

edbtest=#

5.9
按分区列值查询,值使用了子查询,用等号走全表扫描,用in走分区表索引扫描
edbtest=# explain select * from test where id in(select 1 from dual);
                                              QUERY PLAN                                              
-------------------------------------------------------------------------------------------------------
 Nested Loop Semi Join  (cost=0.00..8.31 rows=2 width=47)
   ->  Append  (cost=0.00..7.27 rows=2 width=47)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: (id = 1)
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = 1)
   ->  Materialize  (cost=0.00..1.01 rows=1 width=0)
         ->  Seq Scan on dual  (cost=0.00..1.01 rows=1 width=0)
(8 rows)

edbtest=# explain select * from test where id =(select 1 from dual);
                                                 QUERY PLAN                                                 
-------------------------------------------------------------------------------------------------------------
 Result  (cost=1.01..22.82 rows=4 width=28)
   InitPlan 1 (returns $0)
     ->  Seq Scan on dual  (cost=0.00..1.01 rows=1 width=0)
   ->  Append  (cost=0.00..21.81 rows=4 width=28)
         ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=86)
               Filter: (id = $0)
         ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = $0)
         ->  Index Scan using test_t1001_2000_pkey on test_t1001_2000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = $0)
         ->  Index Scan using test_t2001_3000_pkey on test_t2001_3000 test  (cost=0.00..7.27 rows=1 width=8)
               Index Cond: (id = $0)
(12 rows)

edbtest=#

5.10
按分区列值 更新,走分区表索引扫描
edbtest=# explain update test set name = 'bbb' where id=99;
                                            QUERY PLAN                                           
--------------------------------------------------------------------------------------------------
 Update on test  (cost=0.00..7.27 rows=2 width=10)
   ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=10)
         Filter: (id = 99)
   ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=10)
         Index Cond: (id = 99)
(5 rows)

5.11
按分区列值 删除,走分区表索引扫描
edbtest=# explain delete from test where id=99;
                                           QUERY PLAN                                           
-------------------------------------------------------------------------------------------------
 Delete on test  (cost=0.00..7.27 rows=2 width=6)
   ->  Seq Scan on test  (cost=0.00..0.00 rows=1 width=6)
         Filter: (id = 99)
   ->  Index Scan using test_t1_1000_pkey on test_t1_1000 test  (cost=0.00..7.27 rows=1 width=6)
         Index Cond: (id = 99)
(5 rows)

edbtest=#
　　-----------------

　　转载请著明出处：
blog.csdn.net/beiigang
beigang.iyunv.com