VLDB中的查询优化

Question

我有一个非常大的数据库，大小大约为2gb。该数据库包含10个表，其中一个表有60多列和13902134行（vt_owner）

Insert into insurance.uninsured_vehicle 
           select a.regn_no,a.chasi_no,a.rto_cd,a.state_cd,a.regn_dt,a.vh_class 
from insurance.vt_owner a 
            where a.regn_no not in(select regn_no from insurance.irda_insurance)

执行此查询需要花费大量时间！目前正在使用btree索引。我想优化这个查询，并希望得到一些关于如何进行查询的建议。使用postgres 9.2

我决定对表进行垂直分区，然后在各个分区上应用索引。

vt_owner表定义

CREATE TABLE insurance.vt_owner
(
  regn_no character varying(10) NOT NULL,
  regn_dt timestamp without time zone,
  purchase_dt timestamp without time zone,
  owner_sr numeric(5,0),
  owner_name character varying(150),
  pan_no character varying(10),
  f_name character varying(150),
  c_add1 character varying(50),
  c_add2 character varying(50),
  c_city character varying(50),
  c_district character varying(30),
  c_pincode character varying(6),
  p_add1 character varying(50),
  p_add2 character varying(50),
  p_city character varying(50),
  p_district character varying(30),
  p_pincode character varying(6),
  owner_cd numeric(5,0),
  owner_cd_desc character varying(50),
  regn_type character varying(1),
  regn_type_desc character varying(50),
  vh_class numeric(5,0),
  vh_class_desc character varying(50),
  chasi_no character varying(30),
  eng_no character varying(30),
  maker numeric(5,0),
  maker_desc character varying(50),
  maker_model character varying(50),
  body_type character varying(3),
  body_type_desc character varying(50),
  no_cyl numeric(2,0),
  hp character varying(10),
  seat_cap numeric(3,0),
  stand_cap numeric(3,0),
  sleeper_cap numeric(2,0),
  unld_wt numeric(9,0),
  ld_wt numeric(9,0),
  fuel numeric(3,0),
  fuel_desc character varying(50),
  color character varying(50),
  manu_mon numeric(2,0),
  manu_yr numeric(4,0),
  fit_dt timestamp without time zone,
  norms numeric(2,0),
  norms_desc character varying(50),
  wheelbase character varying(10),
  cubic_cap character varying(10),
  floor_area numeric(7,3),
  ac_fiitted character(1),
  audio_fiitted character(1),
  video_fiitted character(1),
  vch_purchase_as character(1),
  vch_catg character(3),
  dealer_cd numeric(5,0),
  dealer_cd_desc character varying(50),
  sale_amt numeric(9,0),
  laser_code character varying(10),
  garage_add character varying(50),
  state_cd character varying(2) NOT NULL,
  rto_cd character varying(3) NOT NULL,
  CONSTRAINT vt_owner_pkey PRIMARY KEY (state_cd, rto_cd, regn_no)
)
WITH (
  OIDS=FALSE
);
ALTER TABLE insurance.vt_owner
  OWNER TO postgres;
-- Index: insurance."index_VT_OWNER"
-- DROP INDEX insurance."index_VT_OWNER";
CREATE INDEX "index_VT_OWNER"
  ON insurance.vt_owner
  USING btree
  (regn_no COLLATE pg_catalog."default", chasi_no COLLATE pg_catalog."default", state_cd COLLATE pg_catalog."default", rto_cd COLLATE pg_catalog."default");

我也想过创造八卦。求你了，我需要一些帮助。

Answer 1

是否为insurance.vt_owner.regn_no编制索引？

insurance.irda_insurance.regn_no上有索引吗？

你最近运行过ANALYZE吗？

insurance.uninsured_vehicle上有任何触发器吗？

此查询的性能是否更好？

INSERT INTO insurance.uninsured_vehicle 
SELECT a.regn_no,a.chasi_no,a.rto_cd,a.state_cd,a.regn_dt,a.vh_class 
FROM insurance.vt_owner a 
WHERE NOT EXISTS (SELECT 1 FROM insurance.irda_insurance WHERE regn_no = a.regn_no)