Ruby performance - The low hanging fruit

Ruby PerformanceThe Low Hanging Fruit

Hello, I’m Bruce Werdschinski

Senior Lead Developer, Prezentt !

Spent 5 years as the Technical Director at GTP iCommerce

!

~2.5 years with Ruby, built hundreds of ecommerce websites, loves SEO &

CRO !

Mercedes enthusiast

Agenda

• Introduction to common tools

• Scenarios where we would use those tools

• What else can we do, low hanging fruit wise

From experience, look at common expensive tasks

• Database operations

• Network access

• Incorrect algorithm

• Unnecessary work - Can we cache it?

Ruby benchmark

• Disable Garbage Collection first with: GC.disable

• Part of the Ruby standard library

• I prefer to use it like this: !

time = Benchmark.realtime do # The code I’m testing end

benchmark-ips

• https://github.com/evanphx/benchmark-ips

• Benchmarks a blocks iterations/second.

• For short snippits of code, IPS automatically figures out how many times to run the code to get interesting data

• Provides a standard deviation measure which is very cool

stackprof• https://github.com/tmm1/stackprof

• For Ruby 2.1 and above

• Samples what the CPU is spending time on

• https://github.com/alisnic/stackprof-webnav is a Web UI for stackprof dumps

• https://github.com/quirkey/stackprof-remote is middleware and CLI for fetching and interacting with stackprof

How do we find expensive things?• Experience / Knowledge

• Logs

• Ruby's own benchmark module

• External tools

• benchmark-ips, stackprof, time

• SaaS metrics tools

• Librato, Datadog, Skylight, New Relic, Keen IO

Reduce the impact• Optimise the application

• Optimise the environment

Optimise the Application

• Reduce heavy I/O

• ActiveRecord tweaks

• Caching

Reduce heavy I/OSilly example, but it illustrates the point. Writing to the screen is more expensive than writing to memory !1_000_000.times do puts "Hello world!" end !s = String.new 1_000_000.times do s << "Hello world!\n" end puts s

10.5 seconds vs. 6.5 seconds - But 2.28 times more memory

Reduce heavy I/O• Same applies for database calls,

event processing, etc. One large call is often much faster than many little calls.

• I’ve been playing with Keen IO lately, their demo app is at: http://keen-gem-example.herokuapp.com/

• 50 calls, 1 item each: 20094ms

• 1 call with 50 items: 284ms

ActiveRecord

• Know your SQL

• Check logs to see what ActiveRecord is actually doing

• Lots of tools to help us

ActiveRecord sum example

I want to get a total of all to the orders in the database

Order.sum(&:total)

Order Load (49.3ms) SELECT "orders".* FROM "orders"

Order.sum(:total)

(4.0ms) SELECT SUM("orders"."total") AS sum_id FROM "orders"

ActiveRecord sum example

ActiveRecord sum example

ActiveRecord Tools

• rack-mini-profiler

• RailsPanel

• rails-footnotes

• Bullet

• Peek

rack-mini-profiler• https://github.com/MiniProfiler/

rack-mini-profiler

• Middleware that shows a HTML panel on each page

RailsPanel• https://github.com/dejan/

rails_panel

• Chrome Extension that provides information about the Rails request. Includes Database, rendering and total times.

Memoization

Memoization is a caching technique of storing a computed value to avoid duplicated work by future calls.

1. Perform some work

2. Store the result of that work

3. Use the stored data the next time you need the result of that work

MemoizationA common pattern is using the conditional assignment operator: ||=

def current_user

User.find(session[:user_id])

end

Instead store the result in an instance variable by using:

def current_user

@current_user ||= User.find(session[:user_id])

end

identity_cachehttps://github.com/Shopify/identity_cache

ActiveRecord cache for model objects

Extracted from Shopify

@user = User.fetch(id) instead of @user = User.find(id)

Uses an after_commit hook to expire the object from cache

Rails Cache• Use the Dalli gem as the Rails cache store

def fetch(id)

Rails.cache.fetch("product-#{id}", expires_in: 5.minutes) do

Product.find(id)

end

end

Is Ruby the best tool?

• Sometimes…shock, horror…Ruby might not be the best tool for the job.

• Where Ruby may not be the best fit:

• CPU bound

• Async I/O

Leibniz formula for π1 - 1/3 + 1/5 - 1/7 + 1/9 - 1/11 … = π/4

20 million iterations

CPU bound task

• Ruby: 3.78 seconds

• Rust: 0.238 seconds

Optimise the Environment

• Low hanging fruit, faster server, database, more memory

• Ruby versions

• Network topology

• Caching systems

• Web performance

Web performance

• No point shaving 1/10th of a second off your database access time if you’ve got a 3Mb background image.

• http://gtmetrix.com/

• http://tools.pingdom.com/fpt/

No more low hanging fruit

• “Ruby Under a Microscope” by Pat Shaughnessy

• Utilization Saturation and Errors (USE) Method

Thank You!

Twitter: @bwerdschinski !

Email: [email protected] !

Blog: http://www.werdschinski.com