Performance testing matchers for RSpec
Switch branches/tags
Clone or download


Gem Version Build Status Build status Code Climate Coverage Status Inline docs

Performance testing matchers for RSpec

RSpec::Benchmark uses benchmark-perf for measuring execution time and iterations per second and benchmark-trend for asymptotic behaviour estimation.


Integration and unit tests ensure that changing code maintains expected functionality. What is not guaranteed is the code changes impact on library performance. It is easy to refactor your way out of fast to slow code.

If you are new to performance testing you may find Caveats section helpful.



Add this line to your application's Gemfile:

gem 'rspec-benchmark'

And then execute:

$ bundle

Or install it yourself as:

$ gem install rspec-benchmark

1. Usage

For matchers to be available globally, in spec_helper.rb do:

require 'rspec-benchmark'

RSpec.configure do |config|
  config.include RSpec::Benchmark::Matchers

This will add the following matchers:

  • perform_under to see how fast your code runs
  • perform_at_least to see how many iteration per second your code can do
  • perform_(faster|slower)_than to compare implementations
  • perform_(constant|linear|logarithmic|power|exponential) to see how your code scales with time

that will help you express expected performance benchmark for an evaluted code.

Alternatively, you can add matchers for particular example:

RSpec.describe "Performance testing" do
  include RSpec::Benchmark::Matchers

Then you're good to start setting performance expectations:

expect {
}.to perform_under(6).ms

1.1 Timing

The perform_under matcher answers the question of how long does it take to perform a given block of code on average. The measurements are taken executing the block of code in a child process for accurent cpu times.

expect { ... }.to perform_under(0.01).sec

All measurements are assumed to be expressed as seconds. However, you can also provide time in ms, us and ns. The equivalent example in ms would be:

expect { ... }.to perform_under(10).ms
expect { ... }.to perform_under(10000).us

by default the above code will be sampled only once but you can change this by using the sample matcher like so:

expect { ... }.to perform_under(0.01).sample(10) # repeats measurements 10 times

For extra expressiveness you can use times:

expect { ... }.to perform_under(0.01).sample(10).times

You can also use warmup matcher that can run your code before the actual samples are taken to reduce erratic execution times.

For example, you can execute code twice before you take 10 actual measurements:

expect { ... }.to perform_under(0.01).sec.warmup(2).times.sample(10).times

1.2 Iterations

The perform_at_least matcher allows you to establish performance benchmark of how many iterations per second a given block of code should perform. For example, to expect a given code to perform at least 10K iterations per second do:

expect { ... }.to perform_at_least(10000).ips

The ips part is optional but its usage clarifies the intent.

The performance timining of this matcher can be tweaked using the within and warmup matchers. These are expressed as seconds.

By default within matcher is set to 0.2 second and warmup matcher to 0.1 respectively. To change how long measurements are taken, for example, to double the time amount do:

expect { ... }.to perform_at_least(10000).within(0.4).warmup(0.2).ips

The higher values for within and warmup the more accurate average readings and more stable tests at the cost of longer test suite overall runtime.

1.3 Comparison

The perform_faster_than and perform_slower_than matchers allow you to test performance of your code compared with other. For example:

expect { ... }.to perform_faster_than { ... }
expect { ... }.to perform_slower_than { ... }

And if you want to compare how much faster or slower your code is do:

expect { ... }.to perform_faster_than { ... }.once
expect { ... }.to perform_faster_than { ... }.twice
expect { ... }.to perform_faster_than { ... }.exactly(5).times
expect { ... }.to perform_faster_than { ... }.at_least(5).times
expect { ... }.to perform_faster_than { ... }.at_most(5).times

expect { ... }.to perform_slower_than { ... }.once
expect { ... }.to perform_slower_than { ... }.twice
expect { ... }.to perform_slower_than { ... }.at_least(5).times
expect { ... }.to perform_slower_than { ... }.at_most(5).times
expect { ... }.to perform_slower_than { ... }.exactly(5).times

The times part is also optional.

The performance timining of each matcher can be tweaked using the within and warmup matchers. These are expressed as seconds. By default within matcher is set to 0.2 and warmup matcher to 0.1 second respectively. To change these matchers values do:

expect { ... }.to perform_faster_than.within(0.4).warmup(0.2) { ... }

The higher values for within and warmup the more accurate average readings and more stable tests at the cost of longer test suite overall runtime.

1.4 Complexity

The perform_constant, perform_linear, perform_logarithmic, perform_power and perform_exponential matchers are useful for estimating the asymptotic behaviour of a given block of code. The most basic way to use the expectations to test how your code scales is to use the matchers:

expect { ... }.to perform_constant
expect { ... }.to perform_linear
expect { ... }.to perform_logarithmic
expect { ... }.to perform_power
expect { ... }.to perform_exponential

However, for the matchers to be of any use you will need to provide the range of inputs on which they will perform measurements using in_range matcher. Each range input together with its corresponding iteration index will be yielded as arguments to the evaluted block.

For example, to create a power range of inputs from 8 to 100_000 do:

expect { |n, i|
}.to perform_linear.in_range(8, 100_000)

This example will generate and yield input n and index i pairs [8, 0], [64, 1], [512, 2], [4K, 3], [32K, 4] and [100K, 5] respectively.

By default the range will be generated using ratio of 8. You can change this using ratio matcher:

expect { |n, i|
}.to perform_linear.in_range(8, 100_000).ratio(2)

The performance of code block is measured only once per range input. You can change this value and in doing so increase stability of your performance test using the sample matcher. For example, to repeat measurements 100 times for each range data input do:

expect { |n, i|
}.to perform_linear.in_range(8, 100_000).ratio(2).sample(100).times

2. Compounding

All the matchers can be used in compound expressions via and/or. For example, if you wish to check if a computation performs under certain time boundry and iterates at least a given number do:

expect {
}.to perform_under(6).ms and perform_at_least(10000).ips

3. Filtering

Usually performance tests are best left for CI or occasional runs that do not affect TDD/BDD cycle.

To achieve isolation you can use RSpec filters to exclude performance tests from regular runs. For example, in spec_helper:

RSpec.config do |config|
  config.filter_run_excluding perf: true

and then in your example group do:

RSpec.describe ..., :perf do

Then you can run groups or examples tagged with perf:

rspec --tag perf

Another option is to simply isolate the performance specs in separate directory such as spec/performance/... and add custom rake task to run them.

4. Caveats

When writing performance tests things to be mindful are:

  • The tests may potentially be flaky thus its best to use sensible boundaries:
    • too strict boundaries may cause false positives, making tests fail
    • too relaxed boundaries may also lead to false positives missing actual performance regressions
  • Generally performance tests will be slow, but you may try to avoid unnecessarily slow tests by choosing smaller maximum value for sampling

If you have any other observations please share them!


  1. Fork it ( )
  2. Create your feature branch (git checkout -b my-new-feature)
  3. Commit your changes (git commit -am 'Add some feature')
  4. Push to the branch (git push origin my-new-feature)
  5. Create a new Pull Request


Copyright (c) 2016-2018 Piotr Murach. See LICENSE for further details.