No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
app
bin
config
db
lib
log
public
spec
vendor/assets
.gitignore
.rspec
.travis.yml
Gemfile
Gemfile.lock
README.md
Rakefile
config.ru

README.md

ActiveRecord Query Optimization

Shh, don't tell anyone... but all the Rails apps you've been building for the past 9 weeks probably have really bad database performance! Why? Because Rails makes it super easy to write code that causes N+1 queries.

N+1 is a performance issue where, in the process of displaying N records on a page, you end up sending N+1 (or more) SQL queries to the database. This is hugely inefficient, since every round-trip from Ruby to Postgres takes time. It's much better to preload or cache the data you know you'll need, so the number of database round-trips is a small constant number regardless of how many records are being displayed.

A typical example might be displaying a list of articles, which each belong to a user, and wanting to display information about each article's user. We make an initial query for articles in the controller using Article.all, and then every time we call .user on an individual article in the view, another SQL query is sent to retrieve that article's user. If we had 20 articles, we'd be making 21 queries.

Rails has two basic means of solving N+1 issues:

There is also a gem called bullet that can (imperfectly) detect N+1 issues for us, and tell us which of the above techniques we need to use.

Setup

This repo is yet another copy of the example branch from the Week 4 Hacker News project. It now includes seed data that will randomly generate a large-ish number of users, articles, comments, and votes. Run rake db:setup to set up the database, and the seed data will be loaded automatically.

Bullet is set up as per the gem's README – see config/environments/development.rb.

Important bullet caveat!

Bullet can only detect missing counter caches when we use the .size method to get the number of records in a collection or association – using .count or .length will not trigger the warning, even if it generates N+1 queries. You should probably be using .size anyway: See the differences here.