Active Record Presentation Feb 12

work.rowanhick.com

How to avoid hanging yourself with Rails Using ActiveRecord right the first time

1

Discussion tonight • Intended for new Rails Developers • People that think Rails is slow • Focus on simple steps to improve

common :has_many performance problems

• Short - 15mins • All links/references up on

http://work.rowanhick.com tomorrow 2

About me • • • • •

New Zealander (not Australian) Product Development Mgr for a startup in Toronto Full time with Rails for 2 years Previously PHP/MySQL for 4 years 6 years Prior QA/BA/PM for Enterprise CAD/ CAM software dev company

3

Disclaimer •

For sake of brevity and understanding, the SQL shown here is cut down to “psuedo sql”

•

This is not an exhaustive in-depth analysis, just meant as a heads up

•

Times were done using ApacheBench through mongrel in production mode

•

ab -n 1000 http://127.0.0.1/orders/test_xxxx

4

ActiveRecord lets you get in trouble far to quick. •

Super easy syntax comes at a cost. @orders = Order.find(:all) @orders.each do |order| puts order.customer.name puts order.customer.country.name end

✴Congratulations, you just overloaded your DB

with (total number of Orders x 2) unnecessary SQL calls

5

What happened there? • One query to get the orders @orders = Order.find(:all) “SELECT * FROM orders”

• For every item in the orders collection customer.name: “SELECT * FROM customers WHERE id = x” customer.country.name: “SELECT * FROM customers WHERE id = y”

6

Systemic Problem in Web development I’ve seen:

-

15 Second page reloads 10000 queries per page

“ language performs really poorly, we’re going to get it redeveloped in ”

7

Atypical root cause • Failure to build application with *real* data • ie “It worked fine on my machine” but the

developer never loaded up 100’000 records to see what would happen

• Using Rake tasks to build realistic data sets • Test, test, test • tail -f log/development.log 8

Faker to the rescue •

in lib/xchain.rake namespace :xchain do desc "Load fake customers" task :load_customers => :environment do require 'Faker' Customer.find(:all, :conditions => "email LIKE('%XCHAIN_ %')").each { |c| c.destroy } 1..300.times do c = Customer.new c.status_id = rand(3) + 1 c.country_id = rand(243) + 1 c.name = Faker::Company.name c.alternate_name = Faker::Company.name c.phone = Faker::PhoneNumber.phone_number c.email = "XCHAIN_"+Faker::Internet.email c.save end end $ rake xchain:load_customers

9

Eager loading • By using :include in .finds you create sql joins • Pull all required records in one query find(:all, :include => [ :customer, :order_lines ])

✓

order.customer, order.order_lines

find(:all, :include => [ { :customer => :country }, :order_lines ])

✓

order.customer order.customer.country order.order_lines

10

Improvement • Let’s start optimising ... @orders = Order.find(:all, :include => {:customers => :country} )

• Resulting SQL ... “SELECT orders.*, countries.* FROM orders LEFT JOIN customers ON ( customers.id = orders.customers_id ) LEFT JOIN countries ON ( countries.id = customers.country_id)

✓ 7.70 req/s

1.4x faster

11

Select only what you need • Using the :select parameter in the find

options, you can limit the columns you are requesting back from the database

• No point grabbing all columns, if you only want :id and :name

Orders.find(:all, :select => ‘orders.id, orders.name’)

12

The last slide was very important • Not using selects is *okay* provided you have very small columns, and never any binary, or large text data

• You can suddenly saturate your DB connection.

• Imagine our Orders table had an Invoice

column on it storing a pdf of the invoice...

13

Oops • Can’t show a benchmark • :select and :include don’t work together !, reverts back to selecting all columns

• Core team for a long time have not included patches to make it work

• One little sentence in ActiveRecord rdoc “Because eager loading generates the SELECT statement too, the :select option is ignored.” 14

‘mrj’ to the rescue • http://dev.rubyonrails.org/attachment/ticket/ 7147/init.5.rb

• Monkey patch to fix select/include problem • Produces much more efficient SQL

15

Updated finder • Now :select and :include playing nice: @orders = Order.find(:all, :select => 'orders.id, orders.created_at, customers.name, countries.name, order_statuses.name', :include => [{:customer[:name] => :country[:name]}, :order_status[:name]], :conditions => conditions, :order => 'order_statuses.sort_order ASC,order_statuses.id ASC, orders.id DESC')

✓15.15 req/s 2.88x faster 16

r8672 change •

http://blog.codefront.net/2008/01/30/living-on-theedge-of-rails-5-better-eager-loading-and-more/

•

The following uses new improved association load (12 req/s) @orders = Order.find(:all, :include => [{:customer => :country}, :order_status] )

•

The following does not @orders = Order.find(:all, :include => [{:customer => :country}, :order_status], :order => ‘order_statuses.sort_order’ )

17

r8672 output... • Here’s the SQL Order Load (0.000837) 100) LIMIT 10

SELECT * FROM `orders` WHERE (order_status_id <

Customer Load (0.000439) SELECT * FROM `customers` WHERE (customers.id IN (2106,2018,1920,2025,2394,2075,2334,2159,1983,2017)) Country Load (0.000324) SELECT * FROM `countries` WHERE (countries.id IN (33,17,56,150,194,90,91,113,80,54)) OrderStatus Load (0.000291) (order_statuses.id IN (10))

SELECT * FROM `order_statuses` WHERE

18

But I want more • Okay, this still isn’t blazing fast. I’m building the next killr web2.0 app

• Forgetabout associations, just load it via SQL, depending on application, makes a huge difference

• Concentrate on commonly used pages 19

Catch 22 • Hard coding SQL is the fastest solution • No construction of SQL, no generation of ActiveRecord associated classes

• If your DB changes, you have to update SQL

‣ Keep SQL with models where possible 20

It ain’t pretty.. but it’s fast

• Find by SQL class order

def self.find_current_orders find_by_sql("SELECT orders.id, orders.created_at, customers.name as customer_name, countries.name as country_name, order_statuses.name as status_name FROM orders LEFT OUTER JOIN `customers` ON `customers`.id = `orders`.customer_id LEFT OUTER JOIN `countries` ON `countries`.id = `customers`.country_id LEFT OUTER JOIN `order_statuses` ON `order_statuses`.id = `orders`.order_status_id WHERE order_status_id < 100 ORDER BY order_statuses.sort_order ASC,order_statuses.id ASC, orders.id DESC") end end

• 28.90 req/s ( 5.49x faster ) 21

And the results find(:all)

5.26 req/s

find(:all, :include)

7.70 req/s

1.4x

find(:all, :select, :in clude)

15.15 req/s

2.88x

find_by_sql()

28.90 req/s

5.49x

22

Don’t forget indexes • 64000 orders OrderStatus.find(:all).each { |os| puts os.orders.count }

• •

Avg 0.61 req/s no indexes EXPLAIN your SQL ALTER TABLE `xchain_test`.`orders` ADD INDEX order_status_idx(`order_status_id`);

•

Avg 23 req/s after index (37x improvment)

23

Avoid .count • It’s damned slow OrderStatus.find(:all).each { |os| puts os.orders.count }

• Add column orders_count + update code OrderStatus.find(:all).each { |os| puts os.orders_count }

✓34 req/s vs 108 req/s

(3x faster)

24

For the speed freaks • Merb - http://merbivore.com • 38.56 req/s - 7x performance improvement • Nearly identical code • Blazingly fast 25

work.rowanhick.com

The End

26