挣扎在Rails的

Question

优化轨道WHERE NOT IN查询这里的查询，因为它是在轨：

User.limit(20). 
    where.not(id: to_skip, number_of_photos: 0). 
    where(age: @user.seeking_age_min..@user.seeking_age_max). 
    tagged_with(@user.seeking_traits, on: :trait, any: true). 
    tagged_with(@user.seeking_gender, on: :trait, any: true).ids 

而这里的EXPLAIN ANALYZE输出。注意id <> ALL(...)部分缩短了。其中大约有10K个ID。

Limit (cost=23.32..5331.16 rows=20 width=1698) (actual time=2237.871..2243.709 rows=20 loops=1) 
    -> Nested Loop Semi Join (cost=23.32..875817.48 rows=3300 width=1698) (actual time=2237.870..2243.701 rows=20 loops=1) 
     -> Merge Semi Join (cost=22.89..857813.95 rows=8311 width=1702) (actual time=463.757..2220.691 rows=1351 loops=1) 
       Merge Cond: (users.id = users_trait_taggings_356a192.taggable_id) 
       -> Index Scan using users_pkey on users (cost=0.29..834951.51 rows=37655 width=1698) (actual time=455.122..2199.322 rows=7866 loops=1) 
        Index Cond: (id IS NOT NULL) 
        Filter: ((number_of_photos <> 0) AND (age >= 18) AND (age <= 99) AND (id <> ALL ('{7066,7065,...,15624,23254}'::integer[]))) 
        Rows Removed by Filter: 7652 
       -> Index Only Scan using taggings_idx on taggings users_trait_taggings_356a192 (cost=0.42..22767.59 rows=11393 width=4) (actual time=0.048..16.009 rows=4554 loops=1) 
        Index Cond: ((tag_id = 2) AND (taggable_type = 'User'::text) AND (context = 'trait'::text)) 
        Heap Fetches: 4554 
     -> Index Scan using index_taggings_on_taggable_id_and_taggable_type_and_context on taggings users_trait_taggings_5df4b2a (cost=0.42..2.16 rows=1 width=4) (actual time=0.016..0.016 rows=0 loops=1351) 
       Index Cond: ((taggable_id = users.id) AND ((taggable_type)::text = 'User'::text) AND ((context)::text = 'trait'::text)) 
       Filter: (tag_id = ANY ('{4,6}'::integer[])) 
       Rows Removed by Filter: 2 
Total runtime: 2243.913 ms 

Complete version here。

Index Scan using users_pkey on users好像有什么问题，索引扫描需要很长时间。即使有上age，number_of_photos索引和id：

add_index "users", ["age"], name: "index_users_on_age", using: :btree 
add_index "users", ["number_of_photos"], name: "index_users_on_number_of_photos", using: :btree 

to_skip是用户ID的数组不要跳过。 A user有很多skips。每个skip有一个partner_id。

所以下载to_skip我做：

to_skip = @user.skips.pluck(:partner_id) 

我试图查询隔离只是：

sql = User.limit(20). 
    where.not(id: to_skip, number_of_photos: 0). 
    where(age: @user.seeking_age_min..@user.seeking_age_max).to_sql 

而且还与讲解分析得到了同样的问题。同样，用户ID列表被剪切：

Limit (cost=0.00..435.34 rows=20 width=1698) (actual time=0.219..4.844 rows=20 loops=1) 
    -> Seq Scan on users (cost=0.00..819629.38 rows=37655 width=1698) (actual time=0.217..4.838 rows=20 loops=1) 
     Filter: ((id IS NOT NULL) AND (number_of_photos <> 0) AND (age >= 18) AND (age <= 99) AND (id <> ALL ('{7066,7065,...,15624,23254}'::integer[]))) 
     Rows Removed by Filter: 6 
Total runtime: 5.044 ms 

Complete version here。

有关如何在rails + postgres中优化此查询的任何想法？

编辑：下面是相关机型：

User model

class User < ActiveRecord::Base 
    acts_as_messageable required: :body, # default [:topic, :body] 
         dependent: :destroy 

    has_many :skips, :dependent => :destroy 

    acts_as_taggable # Alias for acts_as_taggable_on :tags 
    acts_as_taggable_on :seeking_gender, :trait, :seeking_race 
    scope :by_updated_date, -> { 
    order("updated_at DESC") 
    } 
end 

# schema 

create_table "users", force: :cascade do |t| 
    t.string "email", default: "", null: false 
    t.datetime "created_at", null: false 
    t.datetime "updated_at", null: false 
    t.text  "skips", array: true 
    t.integer "number_of_photos", default: 0 
    t.integer "age" 
end 

add_index "users", ["age"], name: "index_users_on_age", using: :btree 
add_index "users", ["email"], name: "index_users_on_email", unique: true, using: :btree 
add_index "users", ["number_of_photos"], name: "index_users_on_number_of_photos", using: :btree 
add_index "users", ["updated_at"], name: "index_users_on_updated_at", order: {"updated_at"=>:desc}, using: :btree 

Skips model

class Skip < ActiveRecord::Base 
    belongs_to :user 
end 

# schema 

create_table "skips", force: :cascade do |t| 
    t.integer "user_id" 
    t.integer "partner_id" 
    t.datetime "created_at", null: false 
    t.datetime "updated_at", null: false 
end 

add_index "skips", ["partner_id"], name: "index_skips_on_partner_id", using: :btree 
add_index "skips", ["user_id"], name: "index_skips_on_user_id", using: :btree 

Answer 1

速度问题可能是由于在to_skip（约IDS的一长串60Kb）作为数组传入。那么解决方案就是将其重写为子查询的结果，以便Postgress可以更好地优化查询。

建立时to_skip，请尝试使用select而不是pluck。 pluck返回一个数组，然后传递给主查询。 select反过来返回ActiveRecord::Relation，其中的sql可以包含在主查询中，可能会使其更有效。

to_skip = @user.skips.select(:partner_id) 

在发布型号代码之前，很难提出更具体的建议。我将探索的一般方向是尝试将所有相关步骤合并到单个查询中，以便数据库执行优化。

UPDATE

使用select会是这个样子的活动记录查询（我跳过taggable的东西，因为它似乎不影响性能的多）：

User.limit(20). 
    where.not(id: @user.skips.select(:partner_id), number_of_photos: 0). 
    where(age: 0..25) 

这是SQL查询被执行。请注意子查询如何跳过id：

SELECT "users".* FROM "users" 
    WHERE ("users"."number_of_photos" != 0) 
    AND ("users"."id" NOT IN (
     SELECT "skips"."partner_id" 
     FROM "skips" 
     WHERE "skips"."user_id" = 1 
    )) 
    AND ("users"."age" BETWEEN 0 AND 25) 
    LIMIT 20 

尝试以此方式运行您的查询并查看它如何影响性能。