Greenplum建立表--分布鍵
Greenplum是分布式系統,建立表時需要指定分布鍵(建立表需要CREATEDBA許可權),目的在於將資料平均分布到各個segment。選擇分布鍵非常重要,選擇錯了會導致資料不唯一,更嚴重的是會造成SQL效能急劇下降。
Greenplum有兩種分布策略:
1、hash分布。
Greenplum預設使用hash分布策略。該策略可選一個或者多個列作為分布鍵(distribution key,簡稱DK)。分布鍵做hash演算法來確認資料存放到對應的segment上。相同分布索引值會hash到相同的segment上。表上最好有唯一鍵或者主鍵,這樣能保證資料均衡分不到各個segment上。文法,distributed by。
如果沒有主鍵或者唯一鍵,預設選擇第一列作為分布鍵。增加主鍵
2、隨機(randomly)分布。
資料會被隨機分不到segment上,相同記錄可能會存放在不同的segment上。隨機分布可以保證資料平均,但是Greenplum沒有跨節點的唯一鍵約束資料,所以無法保證資料唯一。基於唯一性和效能考慮,推薦使用hash分布,效能部分會另開一篇文檔詳細介紹。文法,distributed randomly。
一、hash分布鍵
建立表,未指定分布列、分布類型,預設建立hash分布表,把第一列ID欄位作為了分布鍵。
testDB=# create table t_hash(id int,name varchar(50)) distributed by (id);CREATE TABLEtestDB=# testDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | Distributed by: (id)
添加主鍵後,主鍵升級為分布鍵替代了id列。
testDB=# alter table t_hash add primary key (name);NOTICE: updating distribution policy to match new primary keyNOTICE: ALTER TABLE / ADD PRIMARY KEY will create implicit index "t_hash_pkey" for table "t_hash" ALTER TABLEtestDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers --------+-----------------------+----------- id | integer | name | character varying(50) | not nullIndexes: "t_hash_pkey" PRIMARY KEY, btree (name)Distributed by: (name)
驗證hash分布表可實現主鍵或者唯一索引值的唯一性
testDB=# insert into t_hash values(1,'szlsd1');INSERT 0 1testDB=#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,主鍵列上依然能夠建立唯一鍵
testDB=# create unique index u_id on t_hash(name);CREATE INDEXtestDB=#testDB=#testDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers--------+-----------------------+----------- id | integer | name | character varying(50) | not nullIndexes: "t_hash_pkey" PRIMARY KEY, btree (name) "u_id" UNIQUE, btree (name)Distributed by: (name)
但是,非主鍵列無法單獨建立唯一索引,想建立的話必須包含多有分布鍵列
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 INDEXtestDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers--------+-----------------------+----------- id | integer | name | character varying(50) | not nullIndexes: "t_hash_pkey" PRIMARY KEY, btree (name) "uk_id" UNIQUE, btree (id, name)Distributed by: (name)
刪除主鍵後,原hash分布鍵依然不變。
testDB=# alter table t_hash drop constraint t_hash_pkey;ALTER TABLEtestDB=# \d t_hash Table "public.t_hash" Column | Type | Modifiers--------+-----------------------+----------- id | integer | name | character varying(50) | not nullDistributed by: (name)
當分布鍵不是主鍵或者唯一鍵時,我們來驗證分布鍵的相同值落在一個segment的結論。
下面的實驗,name列是分布鍵,我們插入相同的name值,可以看到7條記錄都落在了2號segment節點中。
testDB=# insert into t_hash values(1,'szlsd');INSERT 0 1testDB=# insert into t_hash values(2,'szlsd');INSERT 0 1testDB=# insert into t_hash values(3,'szlsd');INSERT 0 1testDB=# insert into t_hash values(4,'szlsd');INSERT 0 1testDB=# insert into t_hash values(5,'szlsd');INSERT 0 1testDB=# insert into t_hash values(6,'szlsd');INSERT 0 1testDB=#testDB=#testDB=# select gp_segment_id,count(*) from t_hash group by gp_segment_id; gp_segment_id | count---------------+------- 2 | 7(1 row)
二、隨機分布鍵
建立隨機分布表需加distributed randomly關鍵字,具體使用哪列作為分布鍵不得而知。
testDB=# create table t_random(id int ,name varchar(100)) distributed randomly;CREATE TABLEtestDB=#testDB=#testDB=# \d t_random Table "public.t_random" Column | Type | Modifiers--------+------------------------+----------- id | integer | name | character varying(100) |Distributed randomly
驗證主鍵/唯一鍵的唯一性,可以看到隨機分布表不能建立主鍵和唯一鍵
testDB=# alter table t_random add primary key (id,name);ERROR: PRIMARY KEY and DISTRIBUTED RANDOMLY are incompatibletestDB=#testDB=# create unique index uk_r_id on t_random(id);ERROR: UNIQUE and DISTRIBUTED RANDOMLY are incompatibletestDB=#
從實驗中可以看出無法實現資料的唯一性。並且,資料插入隨機分布表,並不是輪詢插入,實驗中共有3個segment,但是在1號插入3條記錄,在2號segment節點插入2條記錄後,才在0號segment中插入資料。隨機分布表如何?資料平均分配不得而知。這個實驗也驗證了隨機分布表的相同值分布在不同segment的結論。
testDB=# insert into t_random values(1,'szlsd3');INSERT 0 1testDB=# 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 1testDB=# 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 1testDB=# 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 1testDB=# 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 1testDB=# 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 1testDB=# 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繼承原表分布鍵
Greenplum中有兩種CTAS文法,無論哪種文法,都預設繼承原表的分布鍵。但是,不會繼承表的一些特殊屬性,如主鍵、唯一鍵、APPENDONLY、COMPRESSTYPE(壓縮)等。
testDB=# \d t_hash; Table "public.t_hash" Column | Type | Modifiers--------+-----------------------+----------- id | integer | name | character varying(50) | not nullIndexes: "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 0testDB=# \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 tableCREATE TABLEtestDB=# \d t_hash_2 Table "public.t_hash_2" Column | Type | Modifiers--------+-----------------------+----------- id | integer | name | character varying(50) | not nullDistributed by: (name)
如果CTAS建立表改變分布鍵,加上distributed by即可。
testDB=# create table t_hash_3 as select * from t_hash distributed by (id);SELECT 0testDB=#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 TABLEtestDB=#testDB=# \d t_hash4Did 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 nullDistributed by: (id)
CTAS時,randomly隨機分布鍵要特別注意,一定要加上distributed randomly,不然原表是hash分布鍵,CTAS新表則是隨機分布鍵。
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 7testDB=#testDB=# \d t_random_2 Table "public.t_random_2" Column | Type | Modifiers--------+------------------------+----------- id | integer | name | character varying(100) |Distributed randomly
參考:
《Greenplum公司專屬應用程式實戰》
《Greenplum4.2.2管理員指南》
以上就是Greenplum建立表--分布鍵_PHP教程的內容,更多相關內容請關注topic.alibabacloud.com(www.php.cn)!