Greenplum创建表--分布键
Greenplum是分布式系统,创建表时需要指定分布键(创建表需要CREATEDBA权限),目的在于将数据平均分布到各个segment。选择分布键非常重要,选择错了会导致数据不唯一,更严重的是会造成SQL性能急剧下降。
Greenplum有两种分布策略:
1、hash分布。
Greenplum默认使用hash分布策略。该策略可选一个或者多个列作为分布键(distribution key,简称DK)。分布键做hash算法来确认数据存放到对应的segment上。相同分布键值会hash到相同的segment上。表上最好有唯一键或者主键,这样能保证数据均衡分不到各个segment上。语法,distributed by。
如果没有主键或者唯一键,默认选择第一列作为分布键。增加主键
2、随机(randomly)分布。
数据会被随机分不到segment上,相同记录可能会存放在不同的segment上。随机分布可以保证数据平均,但是Greenplum没有跨节点的唯一键约束数据,所以无法保证数据唯一。基于唯一性和性能考虑,推荐使用hash分布,性能部分会另开一篇文档详细介绍。语法,distributed randomly。
一、hash分布键
创建表,未指定分布列、分布类型,默认创建hash分布表,把第一列ID字段作为了分布键。
testDB=# create table t_hash(id int,name varchar(50)) distributed by (id); CREATE TABLE testDB=# testDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | Distributed by: (id)
添加主键后,主键升级为分布键替代了id列。
testDB=# alter table t_hash add primary key (name); NOTICE: updating distribution policy to match new primary key NOTICE: ALTER TABLE / ADD PRIMARY KEY will create implicit index "t_hash_pkey" for table "t_hash" ALTER TABLE testDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not null Indexes: "t_hash_pkey" PRIMARY KEY, btree (name) Distributed by: (name)
验证hash分布表可实现主键或者唯一键值的唯一性
testDB=# insert into t_hash values(1,'szlsd1'); INSERT 0 1 testDB=# testDB=# insert into t_hash values(2,'szlsd1'); ERROR: duplicate key violates unique constraint "t_hash_pkey"(seg2 gp-s3:40000 pid=3855)
另外,主键列上依然能够创建唯一键
testDB=# create unique index u_id on t_hash(name); CREATE INDEX testDB=# testDB=# testDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not null Indexes: "t_hash_pkey" PRIMARY KEY, btree (name) "u_id" UNIQUE, btree (name) Distributed by: (name)
但是,非主键列无法单独创建唯一索引,想创建的话必须包含多有分布键列
testDB=# create unique index uk_id on t_hash(id); ERROR: UNIQUE index must contain all columns in the distribution key of relation "t_hash" testDB=# create unique index uk_id on t_hash(id,name); CREATE INDEX testDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not null Indexes: "t_hash_pkey" PRIMARY KEY, btree (name) "uk_id" UNIQUE, btree (id, name) Distributed by: (name)
删除主键后,原hash分布键依然不变。
testDB=# alter table t_hash drop constraint t_hash_pkey; ALTER TABLE testDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not null Distributed by: (name)
当分布键不是主键或者唯一键时,我们来验证分布键的相同值落在一个segment的结论。
下面的实验,name列是分布键,我们插入相同的name值,可以看到7条记录都落在了2号segment节点中。
testDB=# insert into t_hash values(1,'szlsd'); INSERT 0 1 testDB=# insert into t_hash values(2,'szlsd'); INSERT 0 1 testDB=# insert into t_hash values(3,'szlsd'); INSERT 0 1 testDB=# insert into t_hash values(4,'szlsd'); INSERT 0 1 testDB=# insert into t_hash values(5,'szlsd'); INSERT 0 1 testDB=# insert into t_hash values(6,'szlsd'); INSERT 0 1 testDB=# testDB=# testDB=# select gp_segment_id,count(*) from t_hash group by gp_segment_id; gp_segment_id | count ---------------+------- 2 | 7 (1 row)
二、随机分布键
创建随机分布表需加distributed randomly关键字,具体使用哪列作为分布键不得而知。
testDB=# create table t_random(id int ,name varchar(100)) distributed randomly; CREATE TABLE testDB=# testDB=# testDB=# \d t_random Table "public.t_random" Column | Type | Modifiers --------+------------------------+----------- id | integer | name | character varying(100) | Distributed randomly
验证主键/唯一键的唯一性,可以看到随机分布表不能创建主键和唯一键
testDB=# alter table t_random add primary key (id,name); ERROR: PRIMARY KEY and DISTRIBUTED RANDOMLY are incompatible testDB=# testDB=# create unique index uk_r_id on t_random(id); ERROR: UNIQUE and DISTRIBUTED RANDOMLY are incompatible testDB=#
从实验中可以看出无法实现数据的唯一性。并且,数据插入随机分布表,并不是轮询插入,实验中共有3个segment,但是在1号插入3条记录,在2号segment节点插入2条记录后,才在0号segment中插入数据。随机分布表如何实现数据平均分配不得而知。这个实验也验证了随机分布表的相同值分布在不同segment的结论。
testDB=# insert into t_random values(1,'szlsd3'); INSERT 0 1 testDB=# select gp_segment_id,count(*) from t_random group by gp_segment_id; gp_segment_id | count ---------------+------- 1 | 1 (1 row) testDB=# testDB=# insert into t_random values(1,'szlsd3'); INSERT 0 1 testDB=# select gp_segment_id,count(*) from t_random group by gp_segment_id; gp_segment_id | count ---------------+------- 2 | 1 1 | 1 (2 rows) testDB=# insert into t_random values(1,'szlsd3'); INSERT 0 1 testDB=# select gp_segment_id,count(*) from t_random group by gp_segment_id; gp_segment_id | count ---------------+------- 2 | 1 1 | 2 (2 rows) testDB=# insert into t_random values(1,'szlsd3'); INSERT 0 1 testDB=# select gp_segment_id,count(*) from t_random group by gp_segment_id; gp_segment_id | count ---------------+------- 2 | 2 1 | 2 (2 rows) testDB=# insert into t_random values(1,'szlsd3'); INSERT 0 1 testDB=# select gp_segment_id,count(*) from t_random group by gp_segment_id; gp_segment_id | count ---------------+------- 2 | 2 1 | 3 (2 rows) testDB=# insert into t_random values(1,'szlsd3'); INSERT 0 1 testDB=# select gp_segment_id,count(*) from t_random group by gp_segment_id; gp_segment_id | count ---------------+------- 2 | 2 1 | 3 0 | 1 (3 rows)
三、CTAS继承原表分布键
Greenplum中有两种CTAS语法,无论哪种语法,都默认继承原表的分布键。但是,不会继承表的一些特殊属性,如主键、唯一键、APPENDONLY、COMPRESSTYPE(压缩)等。
testDB=# \d t_hash; Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not null Indexes: "t_hash_pkey" PRIMARY KEY, btree (name) "uk_id" UNIQUE, btree (id, name) Distributed by: (name) testDB=# testDB=# testDB=# create table t_hash_1 as select * from t_hash; NOTICE: Table doesn't have 'DISTRIBUTED BY' clause -- Using column(s) named 'name' as the Greenplum Database data distribution key for this table. HINT: The 'DISTRIBUTED BY' clause determines the distribution of data. Make sure column(s) chosen are the optimal data distribution key to minimize skew. SELECT 0 testDB=# \d t_hash_1 Table "public.t_hash_1" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | Distributed by: (name) testDB=# testDB=# create table t_hash_2 (like t_hash); NOTICE: Table doesn't have 'distributed by' clause, defaulting to distribution columns from LIKE table CREATE TABLE testDB=# \d t_hash_2 Table "public.t_hash_2" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not null Distributed by: (name)
如果CTAS创建表改变分布键,加上distributed by即可。
testDB=# create table t_hash_3 as select * from t_hash distributed by (id); SELECT 0 testDB=# testDB=# \d t_hash_3 Table "public.t_hash_3" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | Distributed by: (id) testDB=# testDB=# testDB=# create table t_hash_4 (like t_hash) distributed by (id); CREATE TABLE testDB=# testDB=# \d t_hash4 Did not find any relation named "t_hash4". testDB=# \d t_hash_4 Table "public.t_hash_4" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not null Distributed by: (id)
CTAS时,randomly随机分布键要特别注意,一定要加上distributed randomly,不然原表是hash分布键,CTAS新表则是随机分布键。
testDB=# \d t_random Table "public.t_random" Column | Type | Modifiers --------+------------------------+----------- id | integer | name | character varying(100) | Distributed randomly testDB=# testDB=# \d t_random_1 Table "public.t_random_1" Column | Type | Modifiers --------+------------------------+----------- id | integer | name | character varying(100) | Distributed by: (id)
testDB=# create table t_random_2 as select * from t_random distributed randomly; SELECT 7 testDB=# testDB=# \d t_random_2 Table "public.t_random_2" Column | Type | Modifiers --------+------------------------+----------- id | integer | name | character varying(100) | Distributed randomly
参考:
《Greenplum企业应用实战》
《Greenplum4.2.2管理员指南》
以上就是Greenplum创建表--分布键_PHP教程的内容,更多相关内容请关注PHP中文网(www.php.cn)!