GIS附近查找性能优化 - PostGIS long lat geometry distance search tuning using gist knn function
背景
空间数据中对临近点的检索使用非常常见, 例如以经纬度为坐标点, 检索离这个点1公里范围内的其他点的信息.
最近有网友问到这样的问题，如何优化呢.
ps 现在的版本可以直接支持，不需要使用子查询来支持了。
正文
本文将以Postgis为例, 举一个简单的例子, 利用gist 索引加速检索.
测试表 :
create table cust_jw
(
dz varchar(300),
jwd geometry
);
测试数据, 取自经纬度信息网站.
insert into cust_jw values ('杭州', ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163));
insert into cust_jw values ('北京', ST_Transform(ST_GeomFromText('POINT(116.46 39.92)', 4326), 2163));
insert into cust_jw values ('南京', ST_Transform(ST_GeomFromText('POINT(118.78 32.04)', 4326), 2163));
insert into cust_jw values ('南宁', ST_Transform(ST_GeomFromText('POINT(108.33 22.84)', 4326), 2163));
insert into cust_jw values ('贵阳', ST_Transform(ST_GeomFromText('POINT(106.71 26.57)', 4326), 2163));
insert into cust_jw values ('南昌', ST_Transform(ST_GeomFromText('POINT(115.89 28.68)', 4326), 2163));
insert into cust_jw values ('余杭', ST_Transform(ST_GeomFromText('POINT(120.3 30.43)', 4326), 2163));
创建gist索引 :
create index idx_cust_jw_1 on cust_jw using gist(jwd);
这个索引方法支持包含<->两个几何类型的距离排序和&&两个几何类型相交.
详见pg_amop , pg_am, pg_operator, pg_opfamily等系统表.
以下SQL查出北京到杭州的直线距离, 单位米 :
SELECT ST_Distance(
ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163),
ST_Transform(ST_GeomFromText('POINT(116.46 39.92)', 4326), 2163)
);
st_distance
------------------
1256521.71432098
(1 row)
以下SQL 查出表中距离ST_Transform(ST_GeomFromText(‘POINT(120.19 30.26)’, 4326), 2163)这个点20公里的坐标.
函数使用方法参考postgis手册.
digoal=# select *,ST_Distance(jwd, ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163)) from cust_jw where jwd && ST_Buffer(ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163), 20000, 10);
dz | jwd | st_distance
------+----------------------------------------------------+------------------
杭州 | 0101000020730800004C94087D5D4F54C173AA7759E8FB5D41 | 0
余杭 | 0101000020730800000E6E5A20494854C121FC688DA9EF5D41 | 14483.9823187612
(2 rows)
Time: 1.335 ms
前面已经说了, 这个索引访问方法支持&&操作符, <->操作符.
digoal=# explain select *,ST_Distance(jwd, ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163)) from cust_jw where jwd && ST_Buffer(ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163), 20000, 10);
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------
---------------------------------------------------------------
Index Scan using idx_cust_jw_1 on cust_jw (cost=0.14..3.41 rows=1 width=548)
Index Cond: (jwd && '01030000207308000001000000290000004C94087DD53B54C173AA7759E8FB5D411122F50B133C54C154295A2DDAF85D41D751B134CA
3C54C1F4F2B643DFF55D41B6BBAE74F63D54C10FB6A0650AF35D41CDDC4767903F54C1D331586C6DF05D4124855AF48D4154C14B9BC9D018EE5D41AC1BE98FE24354
C1F4F2B6431BEC5D41E89F31897F4654C1DDD11D5181EA5D41CDDC4767544954C1FE67201155E95D412D13EB504F4C54C1383864E89DE85D414C94087D5D4F54C173
AA775960E85D416B1526A96B5254C1383864E89DE85D41CB4BC992665554C1FE67201155E95D41B088DF703B5854C1DDD11D5181EA5D41EC0C286AD85A54C1F4F2B6
431BEC5D4174A3B6052D5D54C14B9BC9D018EE5D41CB4BC9922A5F54C1D331586C6DF05D41E26C6285C46054C10FB6A0650AF35D41C1D65FC5F06154C1F4F2B643DF
F55D4187061CEEA76254C154295A2DDAF85D414C94087DE56254C173AA7759E8FB5D4187061CEEA76254C1922B9585F6FE5D41C1D65FC5F06154C1F261386FF1015E
41E26C6285C46054C1D79E4E4DC6045E41CB4BC9922A5F54C11323974663075E4174A3B6052D5D54C19BB925E2B7095E41EC0C286AD85A54C1F261386FB50B5E41B0
88DF703B5854C10983D1614F0D5E41CB4BC992665554C1E8ECCEA17B0E5E416B1526A96B5254C1AE1C8BCA320F5E414C94087D5D4F54C173AA7759700F5E412D13EB
504F4C54C1AE1C8BCA320F5E41CDDC4767544954C1E8ECCEA17B0E5E41E89F31897F4654C10983D1614F0D5E41AC1BE98FE24354C1F261386FB50B5E4124855AF48D
4154C19BB925E2B7095E41CDDC4767903F54C11323974663075E41B6BBAE74F63D54C1D79E4E4DC6045E41D751B134CA3C54C1F261386FF1015E411122F50B133C54
C1922B9585F6FE5D414C94087DD53B54C173AA7759E8FB5D41'::geometry)
(2 rows)
Time: 1.218 ms
以下SQL 按距离排序.
digoal=# select *,ST_Distance(jwd, ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163)) from cust_jw order by jwd <-> ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163);
dz | jwd | st_distance
------+----------------------------------------------------+------------------
杭州 | 0101000020730800004C94087D5D4F54C173AA7759E8FB5D41 | 0
余杭 | 0101000020730800000E6E5A20494854C121FC688DA9EF5D41 | 14483.9823187612
南京 | 0101000020730800000FFE5AD1D62653C16F4F972A10635E41 | 321491.591341196
南昌 | 010100002073080000B2744BA1FE5253C10975D1494AA25F41 | 503843.306221247
北京 | 0101000020730800006EBBB0F1AB0E4FC17207C71D44525E41 | 1256521.71432098
南宁 | 01010000207308000030806B3882F451C18E3F38DCBB686141 | 1409624.7420143
贵阳 | 01010000207308000082EA89026EE14FC1D6A3AD6E9E786141 | 1732521.31784296
(7 rows)
Time: 0.598 ms
通过以下方法强制排序走索引 :
digoal=# set enable_seqscan=off;
SET
Time: 0.109 ms
digoal=# explain select *,ST_Distance(jwd, ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163)) from cust_jw order by jwd <-> ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163);
QUERY PLAN
--------------------------------------------------------------------------------------
Index Scan using idx_cust_jw_1 on cust_jw (cost=0.14..54.44 rows=140 width=548)
Order By: (jwd <-> '0101000020730800004C94087D5D4F54C173AA7759E8FB5D41'::geometry)
(2 rows)
以下为进一步的优化, 如果点比较密集的话, 这种方法比较好.
digoal=# select * from (select *,ST_Distance(jwd, ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163)) AS dist from cust_jw order by jwd <-> ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163) limit 1000) t where dist<15000;
dz | jwd | dist
------+----------------------------------------------------+------------------
杭州 | 0101000020730800004C94087D5D4F54C173AA7759E8FB5D41 | 0
余杭 | 0101000020730800000E6E5A20494854C121FC688DA9EF5D41 | 14483.9823187612
(2 rows)
Time: 0.634 ms
极致优化
进一步优化, 使用游标, 可以将数据扫描降到极限. (前提是for循环中的sql order by使用了索引)
digoal=# do language plpgsql $$
declare
v_rec record;
v_limit int := 1000;
begin
set local enable_seqscan=off; -- 强制索引, 因为扫描行数够就退出.
for v_rec in select *,ST_Distance(jwd, ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163)) AS dist from cust_jw order by jwd <-> ST_Transform(ST_GeomFromText('POINT(120.19 30.26)', 4326), 2163) loop
if v_limit <=0 then
raise notice '已经取足数据';
return;
end if;
if v_rec.dist > 20000 then
raise notice '满足条件的点已输出完毕';
return;
else
raise notice 'do someting, v_rec:%', v_rec;
end if;
v_limit := v_limit -1;
end loop;
end;
$$;
NOTICE: do someting, v_rec:(杭州,0101000020730800004C94087D5D4F54C173AA7759E8FB5D41,0)
NOTICE: do someting, v_rec:(余杭,0101000020730800000E6E5A20494854C121FC688DA9EF5D41,14483.9823187612)
NOTICE: 满足条件的点已输出完毕
DO
使用这种方法最多扫描比需求结果多1行.
函数化
select * from
(
select *,
ST_Distance (ST_Transform ($1, 26986), ST_Transform (jwd, 26986) ) as dist
from cust_jw
where dz='杭州'
order by ST_Transform (pos, 26986) <-> ST_Transform ($1, 26986) limit 200
) t
where
distance2Center <= 2000.0;
create or replace function ff(geometry, float8, int) returns setof record as $$
declare
v_rec record;
v_limit int := $3;
begin
set local enable_seqscan=off; -- 强制索引, 扫描行数够就退出.
for v_rec in
select *,
ST_Distance ( ST_Transform ($1, 26986), ST_Transform (jwd, 26986) ) as dist
from cust_jw
order by ST_Transform (jwd, 26986) <-> ST_Transform ($1, 26986)
loop
if v_limit <=0 then
raise notice '已经取足数据';
return;
end if;
if v_rec.dz='杭州' and v_rec.dist > $2 then
raise notice '满足条件的点已输出完毕';
return;
elsif v_rec.dz='杭州' then
raise notice 'do someting, v_rec:%', v_rec;
return next v_rec;
else
NULL;
end if;
v_limit := v_limit -1;
end loop;
end;
$$ language plpgsql strict volatile;
select * from ff(ST_GeomFromText ('POINT(114.111618652344 28.332331814237)', 4326),2000.0,1) as t(dz varchar,jwd geometry,dist float8);
SRID
如果需要转换SRID，那么请使用表达式索引，例如ST_Transform (pos, 26986).
参考
1. http://www.ximizi.com/jingweidu.php
2. http://postgis.net/docs/manual-2.0/ST_Distance_Sphere.html
3. http://postgis.net/docs/manual-2.0/ST_Buffer.html
4. http://postgis.net/docs/manual-2.0/ST_Transform.html
5. http://postgis.net/docs/manual-2.0/ST_GeomFromText.html
6. http://postgis.net/docs/manual-2.0/geometry_distance_centroid.html