Postgresql為什麼to_date在分割槽表中無法過濾掉無效分割槽而date可以
阿新 • • 發佈：2019-02-06
t_p為父表，t_p_1,t_p_2是子表 hank=> \d+ t_p Table "hank.t_p" Column | Type | Modifiers | Storage | Stats target | Description --------+-----------------------------+-----------------------------------+----------+--------------+------------- c1 | integer | | plain | | c2 | character varying(20) | | extended | | c3 | timestamp without time zone | | plain | | c4 | character varying(20) | | extended | | c5 | character varying(20) | default 'test'::character varying | extended | | c6 | character varying(20) | default 'hank'::character varying | extended | | Child tables: t_p_1, t_p_2 hank=> \d t_p_2 Table "hank.t_p_2" Column | Type | Modifiers --------+-----------------------------+----------------------------------- c1 | integer | c2 | character varying(20) | c3 | timestamp without time zone | c4 | character varying(20) | c5 | character varying(20) | default 'test'::character varying c6 | character varying(20) | default 'hank'::character varying Check constraints: "t_p_201802" CHECK (c3 >= '2018-02-01'::date AND c3 < '2018-03-01'::date) Inherits: t_p hank=> \d t_p_1 Table "hank.t_p_1" Column | Type | Modifiers --------+-----------------------------+----------------------------------- c1 | integer | c2 | character varying(20) | c3 | timestamp without time zone | c4 | character varying(20) | c5 | character varying(20) | default 'test'::character varying c6 | character varying(20) | default 'hank'::character varying Check constraints: "t_p_201801" CHECK (c3 >= '2018-01-01'::date AND c3 < '2018-02-01'::date) Inherits: t_p 檢視分割槽篩選引數： hank=> show constraint_exclusion ; constraint_exclusion ---------------------- partition (1 row) 如果是to_date無法裁剪,掃描了父表和所有子表 hank=> explain select * from t_p where c3>to_date('2018-02-02','yyyy-mm-dd'); QUERY PLAN -------------------------------------------------------------------------------- Gather (cost=1000.00..76112.02 rows=3 width=208) Workers Planned: 3 -> Append (cost=0.00..75111.72 rows=3 width=208) -> Parallel Seq Scan on t_p (cost=0.00..0.00 rows=1 width=244) Filter: (c3 > to_date('2018-02-02'::text, 'yyyy-mm-dd'::text)) -> Parallel Seq Scan on t_p_1 (cost=0.00..75098.46 rows=1 width=190) Filter: (c3 > to_date('2018-02-02'::text, 'yyyy-mm-dd'::text)) -> Parallel Seq Scan on t_p_2 (cost=0.00..13.26 rows=1 width=190) Filter: (c3 > to_date('2018-02-02'::text, 'yyyy-mm-dd'::text)) (9 rows) date可以裁剪，過濾掉無關分割槽，只掃描父表和t_p_2,達到優化效果 hank=> explain select * from t_p where c3>date'2018-02-02'; QUERY PLAN -------------------------------------------------------------- Append (cost=0.00..14.62 rows=2 width=217) -> Seq Scan on t_p (cost=0.00..0.00 rows=1 width=244) Filter: (c3 > '2018-02-02'::date) -> Seq Scan on t_p_2 (cost=0.00..14.62 rows=1 width=190) Filter: (c3 > '2018-02-02'::date) (5 rows) 這裡to_date為什麼不能過濾掉無效分割槽呢，而date可以，因為過濾分割槽的首要條件是要immutable函式，to_date是stable函式，所以無效 檢視to_date的穩定性 這裡s是stable，i是immutable hank=> select proname,provolatile,proargtypes from pg_proc where proname='to_date'; proname | provolatile | proargtypes ---------+-------------+------------- to_date | s | 25 25 (1 row) hank=> select oid,typname from pg_type where oid in (25); oid | typname -----+--------- 25 | text (1 row) 檢視date穩定性 hank=> select proname,provolatile,proargtypes from pg_proc where proname='date'; proname | provolatile | proargtypes ---------+-------------+------------- date | s | 1184 date | s | 702 date | i | 1114 (3 rows) hank=> select oid,typname from pg_type where oid in (1184,702,1114); oid | typname ------+------------- 702 | abstime 1114 | timestamp 1184 | timestamptz