Database Tests with Rails

@igor_sarcevic ·

We, software developers, like to call ourselves computer scientist, software engineers or data scientists. Hah, what a load of bulls*it!

To do real science, we would need to be able to measure, reproduce and prove our code and methodologies. But I will leave this discussion for an another day…

The reason why I started like this is because I often find myself guessing and hoping for the best instead of doing the hard work, and actually trying to measure and test my ideas.

One particular area, where I often fail to live up to a good scientific practice, is estimating the duration of database actions. But no more! In this article I will explain how I test some of my guesses by creating short lived Rails projects with some sample databases.

Creating Test Projects

I start each test session by creating a test project with Rails, usually with a preset database:

rails new db-test-1 --database=postgresql

Then I follow up with creating an appropriate model that fits my testing needs. For example a Movies table:

rails generate model movies title:string genre:string description:text
rake db:migrate

The finishing step is of course to create some data in the database. I like to make a big number of entries because that way the errors in my logic are easily exposed. For example a million entries are a good test candidate. We can make them in the Rails console.

But because this action usually takes more than a couple of minutes, I like to turn of the Rails logger with the following command before I create the entries to prevent unnecessary overhead:

ActiveRecord::Base.logger.level = 1

To return the origin logger state I will execute the following:

ActiveRecord::Base.logger.level = 0

For the above movies table I would write something like this:

genres = ["action", "commedy", "thriller"]

1000000.times do |index|
  Movies.create :title => "movie-#{index}",
                :genre => genres.sample,
                :description => "#{SecureRandom.uuid * 100}"

  # progress indicator
  puts(index) if index % 1000 == 0
end

This is of course pretty basic if you have used Rails for more than a week or two, but it is a necessary boilerplate. Let’s start with the measurement.

Adding New Columns

For example we can create some migrations that add a new column to the movies table.

rails generate migration AddDirectorToMovies
def change
  add_column :movies, :director, :string
end

On my computer this takes less than a second to finish.

$ rake db:migrate

-- add_column(:movies, :director, :string)
   -> 0.0315s

However if I add a default value on the director column:

def change
  add_column :movies, :director, :default => "Nobody"
end

Than it takes significantly longer:

$ rake db:migrate

-- add_column(:movies, :director, :string, {:default=>"Nobody"})
   -> 24.7215s

This was of course expected, but if you would ask me before this test, I would guess that the time difference is not this big. Measurements ftw!

Adding Database Indexes

Adding indexes to the database can often speed up your queries. But will it always speed them up? Let’s find out with a simple example.

For example let’s try to speed up the following query:

> Movie.where(:genre => "action").count
(157.8ms)  SELECT COUNT(*) FROM "movies"  WHERE "movies"."genre" = 'action'
=> 332953

It isn’t too slow, but if we execute several of these counts in one HTTP request, it can add up quickly.

Let’s add an index to it:

rails generate migration AddIndexToMoviesGenre
def change
  add_index :movies, :genre 
end

Execute it with the following:

$ rake db:migrate
== 20150311212758 AddIndexToMoviesGenre: migrating 
-- add_index(:movies, :genre)
   -> 31.1195s
== 20150311212758 AddIndexToMoviesGenre: migrated (31.1196s)

Whooa! I did not expect more than a couple of seconds for this task. But let’s continue. Will it speed up the query?

> Movie.where(:genre => "action").count
(161.4ms)  SELECT COUNT(*) FROM "movies"  WHERE "movies"."genre" = 'action'
=> 332953

Bummer, it did not help at all! As I later checked, it is pretty much useless to add indexes to column that have a small range of values. That is of course exactly the case with the genre column.

Final words

The above tests are probably a little childish, but some of the things actually surprised me. I hope you liked this article.

Happy hacking!