问题
Using Rails 3.2, Ruby 1.9, geocoder gem. The following query I have in my controller:
# 1500+ms to load
nearby_shops = (current_shop.nearbys(10, :order => 'overall_rating DESC')
.where(:shop_type => shop_type).includes(:photos, :active_property_list =>
:hotel_image_lists).limit(5))
# SQL
Shop Load (4045.6ms) SELECT shops.*, 3958.755864232 * 2 * ASIN(SQRT(POWER(
SIN((36.111927 - shops.lat) * PI() / 180 / 2), 2) + COS(36.111927 * PI()
/ 180) * COS(shops.lat * PI() / 180) * POWER(SIN((-115.171229 - shops.lng)
* PI() / 180 / 2), 2))) AS distance, CAST(DEGREES(ATAN2( RADIANS(shops.lng
- -115.171229), RADIANS(shops.lat - 36.111927))) + 360 AS decimal) % 360 AS
bearing FROM `shops` WHERE `shops`.`shop_type` = 'food' AND (shops.lat BETWEEN
35.96719521688915 AND 36.25665878311085 AND shops.lng BETWEEN
-115.3503819353204 AND -114.9920760646796 AND 3958.755864232 * 2 *
ASIN(SQRT(POWER(SIN((36.111927 - shops.lat) * PI() / 180 / 2), 2) +
COS(36.111927 * PI() / 180) * COS(shops.lat * PI() / 180) * POWER(SIN((
-115.171229 - shops.lng) * PI() / 180 / 2), 2))) <= 10 AND shops.id != 85155)
ORDER BY overall_rating DESC LIMIT 5
- shops.lat) * PI() / 180 / 2), 2) + COS(48.8582411618 * PI() / 180) *
COS(shops.lat * PI() / 180) * POWER(SIN((2.2945044899 - shops.lng) * PI() /
180 / 2), 2))) <= 100 AND shops.id != 517) ORDER BY distance ASC LIMIT 25
OFFSET 0
The problem lies in the nearbys which does calculation on the longitude and latitude. I have already added indexes to the longitude and latitude columns, but it doesn't improve anything.
How can I improve this?
P/S: I removed unrelated conditions which doesn't contribute to the speed of the query.
回答1:
This might not be the most exact or elegant solution, but why not just fudge the math a little. You could write a select clause that did something like this:
.select(["id, name, 69.0975851*sqrt(POWER(shops.lat-?,2) + COS(?*PI()/180)*POWER(shops.lon-?,2)) AS DISTANCE", loc.lat, loc.lat, loc.lon])
You don't really need to use that hairy great circle formula if you're just dealing with small (< 500 mile) distances. You'll end up getting the same ~almost~ answers for a fraction of the computational cost.
来源：https://stackoverflow.com/questions/19143398/ruby-geocoder-nearbys-is-slow