ST_DWITHIN 不使用 GIST 或 BRIN 索引

Question

我正在使用 postgis 函数 ST_DWithin(geography gg1, geography gg2, double precision distance_meters) 来查找点是否在距多边形的指定距离内。我正在运行测试以查看查询需要多长时间，并且解释表明它正在对表运行顺序扫描，而不是使用 BRIN 或 GIST 索引。有人可以建议一种优化它的方法吗？

这是表格 -

table1(incident_geog) 与多边形

CREATE TABLE public.incident_geog
(
    incident_id integer NOT NULL DEFAULT nextval('incident_geog_incident_id_seq'::regclass),
    incident_name character varying(20) COLLATE pg_catalog."default",
    incident_span geography(Polygon,4326),
    CONSTRAINT incident_geog_pkey PRIMARY KEY (incident_id)
)

CREATE INDEX incident_geog_gix
    ON public.incident_geog USING gist
    (incident_span)

带有点和距离的table2(watchzones_geog)

CREATE TABLE public.watchzones_geog
(
    id integer NOT NULL DEFAULT nextval('watchzones_geog_id_seq'::regclass),
    date_created timestamp with time zone DEFAULT now(),
    latitude numeric(10,7) DEFAULT NULL::numeric,
    longitude numeric(10,7) DEFAULT NULL::numeric,
    radius integer,
    "position" geography(Point,4326),
    CONSTRAINT watchzones_geog_pkey PRIMARY KEY (id)
)

CREATE INDEX watchzones_geog_gix
    ON public.watchzones_geog USING gist
    ("position")

带有 st_dwithin 的 Sql

explain select i.incident_id,wz.id from watchzones_geog wz, incident_geog i where ST_DWithin(position,incident_span,wz.radius * 1000);

解释的输出：

Nested Loop  (cost=0.26..418436.69 rows=1 width=8)
-> Seq Scan on watchzones_geog wz  (cost=0.00..13408.01 rows=600001 width=40)
 ->  Index Scan using incident_geog_gix on incident_geog i  (cost=0.26..0.67 rows=1 width=292)
        Index Cond: (incident_span && _st_expand(wz."position", ((wz.radius * 1000))::double precision))
        Filter: ((wz."position" && _st_expand(incident_span, ((wz.radius * 1000))::double precision)) AND _st_dwithin(wz."position", incident_span, ((wz.radius * 1000))::double precision, true))

Answer 1

您的 SQL 实际执行的是在指定距离内为 每个 点找到一些多边形。结果incident_geog.incident_id和watchzones_geog.id一一对应。因为你对每个点进行操作，所以它使用顺序扫描。

我猜你想从多边形开始找点。所以你的SQL需要改表。

explain select i.incident_id,wz.id from incident_geog i, watchzones_geog wz where ST_DWithin(position,incident_span,50);

我们可以看到：

Nested Loop  (cost=0.27..876.00 rows=1 width=16)
   ->  Seq Scan on incident_geog i  (cost=0.00..22.00 rows=1200 width=40)
   ->  Index Scan using watchzones_geog_gix on watchzones_geog wz  (cost=0.27..0.70 rows=1 width=40)
         Index Cond: ("position" && _st_expand(i.incident_span, '50'::double precision))
         Filter: ((i.incident_span && _st_expand("position", '50'::double precision)) AND _st_dwithin("position", i.incident_span, '50'::double precision, true))

因为你操作每一个订单，所以总会有一张表通过顺序扫描遍历所有记录。这两个 SQL 的结果并没有什么不同。关键是你从哪个表开始寻找另一个表的顺序。

也许你可以试试Parallel Query。不要使用Parallel Query：

SET parallel_tuple_cost TO 0;
explain analyze select i.incident_id,wz.id from incident_geog i, watchzones_geog wz where ST_DWithin(position,incident_span,50);

Nested Loop  (cost=0.27..876.00 rows=1 width=16) (actual time=0.002..0.002 rows=0 loops=1)
   ->  Seq Scan on incident_geog i  (cost=0.00..22.00 rows=1200 width=40) (actual time=0.002..0.002 rows=0 loops=1)
   ->  Index Scan using watchzones_geog_gix on watchzones_geog wz  (cost=0.27..0.70 rows=1 width=40) (never executed)
         Index Cond: ("position" && _st_expand(i.incident_span, '50'::double precision))
         Filter: ((i.incident_span && _st_expand("position", '50'::double precision)) AND _st_dwithin("position", i.incident_span, '50'::double precision, true))
 Planning time: 0.125 ms
 Execution time: 0.028 ms

尝试Parallel Query 并将parallel_tuple_cost 设置为2：

SET parallel_tuple_cost TO 2;
explain analyze select i.incident_id,wz.id from incident_geog i, watchzones_geog wz where ST_DWithin(position,incident_span,50);

Nested Loop  (cost=0.27..876.00 rows=1 width=16) (actual time=0.002..0.002 rows=0 loops=1)
       ->  Seq Scan on incident_geog i  (cost=0.00..22.00 rows=1200 width=40) (actual time=0.001..0.001 rows=0 loops=1)
       ->  Index Scan using watchzones_geog_gix on watchzones_geog wz  (cost=0.27..0.70 rows=1 width=40) (never executed)
             Index Cond: ("position" && _st_expand(i.incident_span, '50'::double precision))
             Filter: ((i.incident_span && _st_expand("position", '50'::double precision)) AND _st_dwithin("position", i.incident_span, '50'::double precision, true))
     Planning time: 0.103 ms
     Execution time: 0.013 ms

Answer 2

几个一般要点：

1. 使用 IDENTITY COLUMNS 而不是手动设置序列。
2. 您不需要DEFAULT null::，可空列上的默认值始终为null。
3. 确保在加载这两个表后VACUUM ANALAYZE。

4. 不要使用 SQL-89，而是写出你的 INNER JOIN ... ON

   SELECT i.incident_id,wz.id
   FROM watchzones_geog wz
   INNER JOIN incident_geog i
     ON ST_DWithin(wz.position,i.incident_span,50);
   

5. 在您的explain analyze 中，您的查询中有一个wz.radius * 1000，您将半径设为50。它是哪个？如果静态输入半径，查询序列是否会扫描？

6. 如果您没有在表格中使用纬度和经度，请删除这两列。没有理由将它们存储两次。
7. 我不会使用varchar(20)，而是使用text，它更快，因为没有长度检查，而且实现方式相同。