如何为自由文本搜索索引此查询

时间：2012-08-13 19:48:48

标签： postgresql pg-search

亲爱的postgres用户，

简短的故事，我有一个与州相关的地址表（每个地址都有一个州）。我将为搜索运行以下查询：

SELECT "addresses".*, (ts_rank((to_tsvector('simple', coalesce("addresses"."name"::text, '')) || to_tsvector('simple', coalesce(pg_search_85240c410826a2b0e0f0e5.pg_search_442c4ad3183a256248ef8d::text, ''))), (to_tsquery('simple', ''' ' || 'test' || ' ''' || ':*')), 0)) AS pg_search_rank 
FROM "addresses" 
LEFT OUTER JOIN (SELECT "addresses"."id" AS id, string_agg("states"."name"::text, ' ') AS pg_search_442c4ad3183a256248ef8d 
            FROM "addresses" INNER JOIN "states" ON "states"."id" = "addresses"."state_id" GROUP BY "addresses"."id") pg_search_85240c410826a2b0e0f0e5 ON pg_search_85240c410826a2b0e0f0e5.id = "addresses"."id" 
            WHERE (((to_tsvector('simple', coalesce("addresses"."name"::text, '')) || to_tsvector('simple', coalesce(pg_search_85240c410826a2b0e0f0e5.pg_search_442c4ad3183a256248ef8d::text, ''))) @@ (to_tsquery('simple', ''' ' || 'test' || ' ''' || ':*')))) AND (company_id = 2142) 
            ORDER BY pg_search_rank DESC, "addresses"."id" ASC

直接向前，对吗？这在一台好机器和8581地址上大约需要1.226ms。我需要改进这一点，所以我创建了这两个索引

CREATE INDEX index_addresses_search_by_name ON addresses USING gin(to_tsvector('simple', COALESCE((public.addresses.name)::text, '')))
CREATE INDEX index_state_search_by_name ON states USING gin(to_tsvector('simple', COALESCE((public.states.name)::text, '')))

这应该会有所帮助，在地址上创建一个索引，在状态上创建索引，但它不会:(查询速度和以前一样慢， explain 表明索引没有被使用。

请提出建议，

编辑：

以下是解释分析将显示的内容：

"Sort  (cost=11.39..11.40 rows=1 width=4824) (actual time=0.947..0.947 rows=0 loops=1)"
"  Sort Key: (ts_rank((to_tsvector('simple'::regconfig, COALESCE((public.addresses.name)::text, ''::text)) || to_tsvector('simple'::regconfig, COALESCE((string_agg((states.name)::text, ' '::text)), ''::text))), '''test'':*'::tsquery, 0)), public.addresses.id"
"  Sort Method: quicksort  Memory: 17kB"
"  ->  Nested Loop Left Join  (cost=6.19..11.38 rows=1 width=4824) (actual time=0.905..0.905 rows=0 loops=1)"
"        Join Filter: (public.addresses.id = public.addresses.id)"
"        Filter: ((to_tsvector('simple'::regconfig, COALESCE((public.addresses.name)::text, ''::text)) || to_tsvector('simple'::regconfig, COALESCE((string_agg((states.name)::text, ' '::text)), ''::text))) @@ '''test'':*'::tsquery)"
"        ->  Seq Scan on addresses  (cost=0.00..5.14 rows=1 width=4792) (actual time=0.904..0.904 rows=0 loops=1)"
"              Filter: (company_id = 2142)"
"        ->  HashAggregate  (cost=6.19..6.20 rows=1 width=520) (never executed)"
"              ->  Hash Join  (cost=1.02..6.18 rows=1 width=520) (never executed)"
"                    Hash Cond: (public.addresses.state_id = states.id)"
"                    ->  Seq Scan on addresses  (cost=0.00..5.11 rows=11 width=8) (never executed)"
"                    ->  Hash  (cost=1.01..1.01 rows=1 width=520) (never executed)"
"                          ->  Seq Scan on states  (cost=0.00..1.01 rows=1 width=520) (never executed)"
"Total runtime: 1.226 ms"

EDIT2

地址ddl 

--
-- PostgreSQL database dump
--

SET statement_timeout = 0;
SET client_encoding = 'UTF8';
SET standard_conforming_strings = on;
SET check_function_bodies = false;
SET client_min_messages = warning;

SET search_path = public, pg_catalog;

SET default_tablespace = '';

SET default_with_oids = false;

--
-- Name: addresses; Type: TABLE; Schema: public; Owner: trucking; Tablespace: 
--

CREATE TABLE addresses (
    id integer NOT NULL,
    name character varying(255),
    street character varying(255),
    city character varying(255),
    zip_code character varying(255),
    primary_phone character varying(255),
    alternate_phone character varying(255),
    fax character varying(255),
    email character varying(255),
    contact character varying(255),
    company_id integer DEFAULT 0 NOT NULL,
    motor_carrier_number character varying(12),
    state_id integer,
    created_at timestamp without time zone,
    updated_at timestamp without time zone,
    alternate_phone2 character varying(12),
    insurance_expires_on date,
    notes text
);


ALTER TABLE public.addresses OWNER TO trucking;

--
-- Name: addresses_id_seq; Type: SEQUENCE; Schema: public; Owner: trucking
--

CREATE SEQUENCE addresses_id_seq
    START WITH 1
    INCREMENT BY 1
    NO MINVALUE
    NO MAXVALUE
    CACHE 1;


ALTER TABLE public.addresses_id_seq OWNER TO trucking;

--
-- Name: addresses_id_seq; Type: SEQUENCE OWNED BY; Schema: public; Owner: trucking
--

ALTER SEQUENCE addresses_id_seq OWNED BY addresses.id;


--
-- Name: id; Type: DEFAULT; Schema: public; Owner: trucking
--

ALTER TABLE ONLY addresses ALTER COLUMN id SET DEFAULT nextval('addresses_id_seq'::regclass);


--
-- Name: addresses_pkey; Type: CONSTRAINT; Schema: public; Owner: trucking; Tablespace: 
--

ALTER TABLE ONLY addresses
    ADD CONSTRAINT addresses_pkey PRIMARY KEY (id);


--
-- Name: index_addresses_on_company_id; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_addresses_on_company_id ON addresses USING btree (company_id);


--
-- Name: index_addresses_on_state_id; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_addresses_on_state_id ON addresses USING btree (state_id);


--
-- Name: index_addresses_search_by_city; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_addresses_search_by_city ON addresses USING gin (to_tsvector('simple'::regconfig, COALESCE((city)::text, ''::text)));


--
-- Name: index_addresses_search_by_email; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_addresses_search_by_email ON addresses USING gin (to_tsvector('simple'::regconfig, COALESCE((email)::text, ''::text)));


--
-- Name: index_addresses_search_by_name; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_addresses_search_by_name ON addresses USING gin (to_tsvector('simple'::regconfig, COALESCE((name)::text, ''::text)));


--
-- Name: index_addresses_search_by_street; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_addresses_search_by_street ON addresses USING gin (to_tsvector('simple'::regconfig, COALESCE((street)::text, ''::text)));


--
-- PostgreSQL database dump complete
--

国家DDL 

--
-- PostgreSQL database dump
--

SET statement_timeout = 0;
SET client_encoding = 'UTF8';
SET standard_conforming_strings = on;
SET check_function_bodies = false;
SET client_min_messages = warning;

SET search_path = public, pg_catalog;

SET default_tablespace = '';

SET default_with_oids = false;

--
-- Name: states; Type: TABLE; Schema: public; Owner: trucking; Tablespace: 
--

CREATE TABLE states (
    id integer NOT NULL,
    name character varying(255) NOT NULL,
    abbrev character varying(255) NOT NULL,
    country character varying(255)
);


ALTER TABLE public.states OWNER TO trucking;

--
-- Name: states_id_seq; Type: SEQUENCE; Schema: public; Owner: trucking
--

CREATE SEQUENCE states_id_seq
    START WITH 1
    INCREMENT BY 1
    NO MINVALUE
    NO MAXVALUE
    CACHE 1;


ALTER TABLE public.states_id_seq OWNER TO trucking;

--
-- Name: states_id_seq; Type: SEQUENCE OWNED BY; Schema: public; Owner: trucking
--

ALTER SEQUENCE states_id_seq OWNED BY states.id;


--
-- Name: id; Type: DEFAULT; Schema: public; Owner: trucking
--

ALTER TABLE ONLY states ALTER COLUMN id SET DEFAULT nextval('states_id_seq'::regclass);


--
-- Name: states_pkey; Type: CONSTRAINT; Schema: public; Owner: trucking; Tablespace: 
--

ALTER TABLE ONLY states
    ADD CONSTRAINT states_pkey PRIMARY KEY (id);


--
-- Name: index_states_search_by_abbrev; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_states_search_by_abbrev ON states USING gin (to_tsvector('simple'::regconfig, COALESCE((abbrev)::text, ''::text)));


--
-- Name: index_states_search_by_name; Type: INDEX; Schema: public; Owner: trucking; Tablespace: 
--

CREATE INDEX index_states_search_by_name ON states USING gin (to_tsvector('simple'::regconfig, COALESCE((name)::text, ''::text)));


--
-- PostgreSQL database dump complete
--

目前为止最佳重写

SELECT consolidated_address.id, (ts_rank((to_tsvector('simple', coalesce(consolidated_address.name::text, '')) || to_tsvector('simple', coalesce(consolidated_address.state_name::text, ''))), (to_tsquery('simple', ''' ' || 'Gallaway' || ' ''' || ':*')), 0)) AS pg_search_rank 
FROM (
 SELECT "addresses".id,
    "addresses".name,
    string_agg("states".name::text, ' ') as state_name
    FROM addresses
    LEFT OUTER JOIN "states"
    ON "states".id = "addresses".state_id
    GROUP BY "addresses".id) consolidated_address
WHERE 
    (((to_tsvector('simple', coalesce(consolidated_address.name::text, '')) || to_tsvector('simple', coalesce(consolidated_address.state_name::text, ''))) @@ (to_tsquery('simple', ''' ' || 'Gallaway' || ' ''' || ':*'))))

它有点快，但仍然没有使用任何索引

谢谢，

3 个答案:

答案 0 :(得分：1)

在https://github.com/Casecommons/pg_search/issues/51

回复Github的帖子

我是生成此查询的pg_search Ruby gem的作者。

是的，遗憾的是，我不知道如何获取：associated_against查询可以对索引起作用，至少是当前实现的方式。

这是因为：associated_against搜索连接在一起的所有关联记录文本，而不是一次搜索单个记录。

因此，例如，如果你加入了一个标签表，并且有3条记录（“foo”，“bar”和“baz”），那么你会期望搜索“foo baz”来找到它。更容易索引的解决方案只适用于“foo”或“baz”查询，但不适用于“foo baz”，因为如果你知道我的意思，那么单个标签行中没有一行符合这两个术语。

不可能跨多个记录进行索引（至少据我所知）。

也许我们可以在pg_search中找到一个选项来逐个记录进行搜索，这可以使用索引，但不能匹配记录。

答案 1 :(得分：1)

我对你的解释分析输出的读数是你的表太小而无法使用索引。 PostgreSQL上的顺序扫描相对便宜，因为表是作为物理文件加载的，PostgreSQL可以利用操作系统预取等。

请记住，没有计划比从磁盘加载单个页面并按顺序扫描它更快。使用索引只会增加开销。在顺序扫描中没有花费任何重要时间，因此在您拥有真实数据之前不要担心。

值得注意的是，在PostgreSQL上，在拥有数据之前优化查询比解决方案更容易解决问题。我真诚的建议是等待分析查询，直到你有一个真实的数据返回真实数据。然后我们可以确定哪些索引有用。

编辑：你说查询很慢。但该计划显示它在1毫秒内执行。你期望它花多少时间？

答案 2 :(得分：0)

此查询很慢，因为您通过左外连接加入的查询正在使用聚合整个地址表，然后通过company_id进行限制。地址表是否有company_id？尝试将where子句移动到内部查询以限制返回的记录数。

相关问题
最新问题