PostgreSQL DBA(46) - PG Operator classes and families

先前章節(jié)已簡(jiǎn)單提到pg_am,以hash為例介紹了hash index不能支持大于等于,小于等于等相關(guān)操作,我們了解到需要信息來(lái)確認(rèn)訪問(wèn)方法接受哪些數(shù)據(jù)類型和哪些操作符.

[pg12@localhost ~]$ psql -d testdb
psql (12beta1)
Type "help" for help.
testdb=# select * from pg_am;
  oid  |    amname    |      amhandler       | amtype 
-------+--------------+----------------------+--------
     2 | heap         | heap_tableam_handler | t
   403 | btree        | bthandler            | i
   405 | hash         | hashhandler          | i
   783 | gist         | gisthandler          | i
  2742 | gin          | ginhandler           | i
  4000 | spgist       | spghandler           | i
  3580 | brin         | brinhandler          | i
 24597 | blackhole_am | blackhole_am_handler | t
(8 rows)

本節(jié)在此基礎(chǔ)上介紹Operator classes and families.

基礎(chǔ)知識(shí)
PostgreSQL提供了 operator class 和 operator family 兩個(gè)概念來(lái)說(shuō)明訪問(wèn)方法接受哪些數(shù)據(jù)類型和哪些操作符. operator class 包含了索引操作特定數(shù)據(jù)類型的最小操作符集合, operator class 被歸類為 operator family ,也就是或一個(gè)family可能包含多個(gè) operator class .

testdb=# select * from pg_opfamily;
 oid  | opfmethod |        opfname        | opfnamespace | opfowner 
------+-----------+-----------------------+--------------+----------
  397 |       403 | array_ops             |           11 |       10
  627 |       405 | array_ops             |           11 |       10
  423 |       403 | bit_ops               |           11 |       10
  424 |       403 | bool_ops              |           11 |       10
  426 |       403 | bpchar_ops            |           11 |       10
  427 |       405 | bpchar_ops            |           11 |       10
  428 |       403 | bytea_ops             |           11 |       10
...
(107 rows)

上面是系統(tǒng)中存在的所有的opfamily,包括數(shù)組array_ops/位bit_ops/字符串bpchar_ops/整型integer_ops等數(shù)據(jù)類型等op family.比如integer_ops family包括了int8_ops, int4_ops, and int2_ops classes,分別對(duì)應(yīng)bigint, integer, 和 smallint這幾個(gè)類型.
pg_opfamily中的opfmethod字段與pg_am.oid關(guān)聯(lián),比如查詢hash am的opfamily和opclass:

testdb=# select am.amname,opfname, opcname, opcintype::regtype
testdb-# from pg_opclass opc, pg_opfamily opf, pg_am am
testdb-# where opc.opcfamily = opf.oid
testdb-# and opf.opfmethod = am.oid
testdb-# and am.amname = 'hash'
testdb-# order by opf.opfname;
 amname |      opfname       |       opcname       |          opcintype          
--------+--------------------+---------------------+-----------------------------
 hash   | aclitem_ops        | aclitem_ops         | aclitem
 hash   | array_ops          | array_ops           | anyarray
 hash   | bool_ops           | bool_ops            | boolean
 hash   | bpchar_ops         | bpchar_ops          | character
 hash   | bpchar_pattern_ops | bpchar_pattern_ops  | character
 hash   | bytea_ops          | bytea_ops           | bytea
 hash   | char_ops           | char_ops            | "char"
 hash   | cid_ops            | cid_ops             | cid
 hash   | date_ops           | date_ops            | date
 hash   | enum_ops           | enum_ops            | anyenum
 hash   | float_ops          | float8_ops          | double precision
 hash   | float_ops          | float4_ops          | real
 hash   | integer_ops        | int2_ops            | smallint
 hash   | integer_ops        | int8_ops            | bigint
 hash   | integer_ops        | int4_ops            | integer
 hash   | interval_ops       | interval_ops        | interval
 hash   | jsonb_ops          | jsonb_ops           | jsonb
 hash   | macaddr8_ops       | macaddr8_ops        | macaddr8
 hash   | macaddr_ops        | macaddr_ops         | macaddr
 hash   | network_ops        | cidr_ops            | inet
 hash   | network_ops        | inet_ops            | inet
 hash   | numeric_ops        | numeric_ops         | numeric
 hash   | oid_ops            | oid_ops             | oid
 hash   | oidvector_ops      | oidvector_ops       | oidvector
 hash   | pg_lsn_ops         | pg_lsn_ops          | pg_lsn
 hash   | range_ops          | range_ops           | anyrange
 hash   | text_ops           | text_ops            | text
 hash   | text_ops           | name_ops            | name
 hash   | text_ops           | varchar_ops         | text
 hash   | text_pattern_ops   | text_pattern_ops    | text
 hash   | text_pattern_ops   | varchar_pattern_ops | text
 hash   | tid_ops            | tid_ops             | tid
 hash   | time_ops           | time_ops            | time without time zone
 hash   | timestamp_ops      | timestamp_ops       | timestamp without time zone
 hash   | timestamptz_ops    | timestamptz_ops     | timestamp with time zone
 hash   | timetz_ops         | timetz_ops          | time with time zone
 hash   | uuid_ops           | uuid_ops            | uuid
 hash   | xid_ops            | xid_ops             | xid
(38 rows)

可以看到,對(duì)于integer_ops opfamily,可以支持int2_ops/int4_ops/int8_ops這幾類op clas,類型分別是smallint/integer/bigint.
op family可以包括額外的操作符用以比較不同類型的值,之所以歸為同一個(gè)family是因?yàn)樵谑褂胕ndex時(shí),謂詞可以適配不同的數(shù)據(jù)類型(如smallint/integer/bigint等).在大多數(shù)情況下,不需要知道op family和op class,只需要?jiǎng)?chuàng)建索引然后使用就好了.但,可以顯式指定op lcass.

System catalog
下面是op family和op class的相關(guān)系統(tǒng)目錄關(guān)系圖:

通過(guò)上圖,可找出相關(guān)的信息.
找出AM可處理的數(shù)據(jù)類型

testdb=# select am.amname,opcname, opcintype::regtype
testdb-# from pg_opclass opc, pg_am am 
testdb-# where opc.opcmethod = am.oid
testdb-# and am.amname = 'hash'
testdb-# order by opcintype::regtype::text;
 amname |       opcname       |          opcintype          
--------+---------------------+-----------------------------
 hash   | aclitem_ops         | aclitem
 hash   | array_ops           | anyarray
 hash   | enum_ops            | anyenum
 hash   | range_ops           | anyrange
 hash   | int8_ops            | bigint
 hash   | bool_ops            | boolean
 hash   | bytea_ops           | bytea
 hash   | char_ops            | "char"
 hash   | bpchar_pattern_ops  | character
 hash   | bpchar_ops          | character
 hash   | cid_ops             | cid
 hash   | date_ops            | date
 hash   | float8_ops          | double precision
 hash   | cidr_ops            | inet
 hash   | inet_ops            | inet
 hash   | int4_ops            | integer
 hash   | interval_ops        | interval
 hash   | jsonb_ops           | jsonb
 hash   | macaddr_ops         | macaddr
 hash   | macaddr8_ops        | macaddr8
 hash   | name_ops            | name
 hash   | numeric_ops         | numeric
 hash   | oid_ops             | oid
 hash   | oidvector_ops       | oidvector
 hash   | pg_lsn_ops          | pg_lsn
 hash   | float4_ops          | real
 hash   | int2_ops            | smallint
 hash   | text_ops            | text
 hash   | varchar_ops         | text
 hash   | text_pattern_ops    | text
 hash   | varchar_pattern_ops | text
 hash   | tid_ops             | tid
 hash   | timestamp_ops       | timestamp without time zone
 hash   | timestamptz_ops     | timestamp with time zone
 hash   | time_ops            | time without time zone
 hash   | timetz_ops          | time with time zone
 hash   | uuid_ops            | uuid
 hash   | xid_ops             | xid
(38 rows)

哪些op在op class中?(索引訪問(wèn)可用于該操作符的謂詞)

testdb=# select amop.amopopr::regoperator
testdb-# from pg_opclass opc, pg_opfamily opf, pg_am am, pg_amop amop
testdb-# where opc.opcname = 'int8_ops'
testdb-# and opf.oid = opc.opcfamily
testdb-# and am.oid = opf.opfmethod
testdb-# and amop.amopfamily = opc.opcfamily
testdb-# and am.amname = 'hash'
testdb-# and amop.amoplefttype = opc.opcintype;
      amopopr       
--------------------
 =(bigint,bigint)
 =(bigint,smallint)
 =(bigint,integer)
(3 rows)

參考資料
Indexes in PostgreSQL — 2