Testing best practices

Test Design

Testing at GitLab is a first class citizen, not an afterthought. It’s important we consider the design of our tests as we do the design of our features.

When implementing a feature, we think about developing the right capabilities the right way, which helps us narrow our scope to a manageable level. When implementing tests for a feature, we must think about developing the right tests, but then cover all the important ways the test may fail, which can quickly widen our scope to a level that is difficult to manage.

Test heuristics can help solve this problem. They concisely address many of the common ways bugs manifest themselves within our code. When designing our tests, take time to review known test heuristics to inform our test design. We can find some helpful heuristics documented in the Handbook in the Test Engineering section.

Test speed

GitLab has a massive test suite that, without parallelization, can take hours to run. It’s important that we make an effort to write tests that are accurate and effective as well as fast.

Here are some things to keep in mind regarding test performance:

  • double and spy are faster than FactoryBot.build(...)
  • FactoryBot.build(...) and .build_stubbed are faster than .create.
  • Don’t create an object when build, build_stubbed, attributes_for, spy, or double will do. Database persistence is slow!
  • Don’t mark a feature as requiring JavaScript (through :js in RSpec) unless it’s actually required for the test to be valid. Headless browser testing is slow!

RSpec

To run rspec tests:

# run all tests
bundle exec rspec

# run test for path
bundle exec rspec spec/[path]/[to]/[spec].rb

Use guard to continuously monitor for changes and only run matching tests:

bundle exec guard

When using spring and guard together, use SPRING=1 bundle exec guard instead to make use of spring.

General guidelines

  • Use a single, top-level describe ClassName block.
  • Use .method to describe class methods and #method to describe instance methods.
  • Use context to test branching logic.
  • Try to match the ordering of tests to the ordering within the class.
  • Try to follow the Four-Phase Test pattern, using newlines to separate phases.
  • Use Gitlab.config.gitlab.host rather than hard coding 'localhost'
  • Don’t assert against the absolute value of a sequence-generated attribute (see Gotchas).
  • Don’t supply the :each argument to hooks since it’s the default.
  • On before and after hooks, prefer it scoped to :context over :all
  • When using evaluate_script("$('.js-foo').testSomething()") (or execute_script) which acts on a given element, use a Capybara matcher beforehand (e.g. find('.js-foo')) to ensure the element actually exists.
  • Use focus: true to isolate parts of the specs you want to run.
  • Use :aggregate_failures when there is more than one expectation in a test.
  • For empty test description blocks, use specify rather than it do if the test is self-explanatory.

System / Feature tests

Note: Before writing a new system test, please consider not writing one!
  • Feature specs should be named ROLE_ACTION_spec.rb, such as user_changes_password_spec.rb.
  • Use scenario titles that describe the success and failure paths.
  • Avoid scenario titles that add no information, such as “successfully”.
  • Avoid scenario titles that repeat the feature title.
  • Create only the necessary records in the database
  • Test a happy path and a less happy path but that’s it
  • Every other possible path should be tested with Unit or Integration tests
  • Test what’s displayed on the page, not the internals of ActiveRecord models. For instance, if you want to verify that a record was created, add expectations that its attributes are displayed on the page, not that Model.count increased by one.
  • It’s ok to look for DOM elements but don’t abuse it since it makes the tests more brittle

Debugging Capybara

Sometimes you may need to debug Capybara tests by observing browser behavior.

Live debug

You can pause Capybara and view the website on the browser by using the live_debug method in your spec. The current page will be automatically opened in your default browser. You may need to sign in first (the current user’s credentials are displayed in the terminal).

To resume the test run, press any key.

For example:

$ bin/rspec spec/features/auto_deploy_spec.rb:34
Running via Spring preloader in process 8999
Run options: include {:locations=>{"./spec/features/auto_deploy_spec.rb"=>[34]}}

Current example is paused for live debugging
The current user credentials are: user2 / 12345678
Press any key to resume the execution of the example!
Back to the example!
.

Finished in 34.51 seconds (files took 0.76702 seconds to load)
1 example, 0 failures

Note: live_debug only works on JavaScript enabled specs.

Run :js spec in a visible browser

Run the spec with CHROME_HEADLESS=0, e.g.:

CHROME_HEADLESS=0 bundle exec rspec some_spec.rb

The test will go by quickly, but this will give you an idea of what’s happening. Using live_debug with CHROME_HEADLESS=0 pauses the open browser, and does not open the page again. This can be used to debug and inspect elements.

You can also add byebug or binding.pry to pause execution and step through the test.

Screenshots

We use the capybara-screenshot gem to automatically take a screenshot on failure. In CI you can download these files as job artifacts.

Also, you can manually take screenshots at any point in a test by adding the methods below. Be sure to remove them when they are no longer needed! See https://github.com/mattheworiordan/capybara-screenshot#manual-screenshots for more.

Add screenshot_and_save_page in a :js spec to screenshot what Capybara “sees”, and save the page source.

Add screenshot_and_open_image in a :js spec to screenshot what Capybara “sees”, and automatically open the image.

The HTML dumps created by this are missing CSS. This results in them looking very different from the actual application. There is a small hack to add CSS which makes debugging easier.

Fast unit tests

Some classes are well-isolated from Rails and you should be able to test them without the overhead added by the Rails environment and Bundler’s :default group’s gem loading. In these cases, you can require 'fast_spec_helper' instead of require 'spec_helper' in your test file, and your test should run really fast since:

  • Gems loading is skipped
  • Rails app boot is skipped
  • GitLab Shell and Gitaly setup are skipped
  • Test repositories setup are skipped

fast_spec_helper also support autoloading classes that are located inside the lib/ directory. It means that as long as your class / module is using only code from the lib/ directory you will not need to explicitly load any dependencies. fast_spec_helper also loads all ActiveSupport extensions, including core extensions that are commonly used in the Rails environment.

Note that in some cases, you might still have to load some dependencies using require_dependency when a code is using gems or a dependency is not located in lib/.

For example, if you want to test your code that is calling the Gitlab::UntrustedRegexp class, which under the hood uses re2 library, you should either add require_dependency 're2' to files in your library that need re2 gem, to make this requirement explicit, or you can add it to the spec itself, but the former is preferred.

It takes around one second to load tests that are using fast_spec_helper instead of 30+ seconds in case of a regular spec_helper.

let variables

GitLab’s RSpec suite has made extensive use of let(along with it strict, non-lazy version let!) variables to reduce duplication. However, this sometimes comes at the cost of clarity, so we need to set some guidelines for their use going forward:

  • let! variables are preferable to instance variables. let variables are preferable to let! variables. Local variables are preferable to let variables.
  • Use let to reduce duplication throughout an entire spec file.
  • Don’t use let to define variables used by a single test; define them as local variables inside the test’s it block.
  • Don’t define a let variable inside the top-level describe block that’s only used in a more deeply-nested context or describe block. Keep the definition as close as possible to where it’s used.
  • Try to avoid overriding the definition of one let variable with another.
  • Don’t define a let variable that’s only used by the definition of another. Use a helper method instead.
  • let! variables should be used only in case if strict evaluation with defined order is required, otherwise let will suffice. Remember that let is lazy and won’t be evaluated until it is referenced.

Common test setup

In some cases, there is no need to recreate the same object for tests again for each example. For example, a project and a guest of that project is needed to test issues on the same project, one project and user will do for the entire file. This can be achieved by using let_it_be variables and the before_all hook from the test-prof gem.

let_it_be(:project) { create(:project) }
let_it_be(:user) { create(:user) }

before_all do
  project.add_guest(user)
end

This will result in only one Project, User, and ProjectMember created for this context.

let_it_be and before_all are also available within nested contexts. Cleanup after the context is handled automatically using a transaction rollback.

Note that if you modify an object defined inside a let_it_be block, then you will need to reload the object as needed, or specify the reload option to reload for every example.

let_it_be(:project, reload: true) { create(:project) }

You can also specify the refind option as well to completely load a new object.

let_it_be(:project, refind: true) { create(:project) }

set variables

Note: We are incrementally removing set in favour of let_it_be. See the removal issue.

In some cases there is no need to recreate the same object for tests again for each example. For example, a project is needed to test issues on the same project, one project will do for the entire file. This can be achieved by using set in the same way you would use let.

rspec-set only works on ActiveRecord objects, and before new examples it reloads or recreates the model, only if needed. That is, when you changed properties or destroyed the object.

Note that you can’t reference a model defined in a let block in a set block.

Also, set is not supported in :js specs since those don’t use transactions to clean up database state after each example.

Time-sensitive tests

Timecop is available in our Ruby-based tests for verifying things that are time-sensitive. Any test that exercises or verifies something time-sensitive should make use of Timecop to prevent transient test failures.

Example:

it 'is overdue' do
  issue = build(:issue, due_date: Date.tomorrow)

  Timecop.freeze(3.days.from_now) do
    expect(issue).to be_overdue
  end
end

Feature flags in tests

All feature flags are stubbed to be enabled by default in our Ruby-based tests.

To disable a feature flag in a test, use the stub_feature_flags helper. For example, to globally disable the ci_live_trace feature flag in a test:

stub_feature_flags(ci_live_trace: false)

Feature.enabled?(:ci_live_trace) # => false

If you wish to set up a test where a feature flag is disabled for some actors and not others, you can specify this in options passed to the helper. For example, to disable the ci_live_trace feature flag for a specifc project:

project1, project2 = build_list(:project, 2)

# Feature will only be disabled for project1
stub_feature_flags(ci_live_trace: { enabled: false, thing: project1 })

Feature.enabled?(:ci_live_trace, project1) # => false
Feature.enabled?(:ci_live_trace, project2) # => true

Pristine test environments

The code exercised by a single GitLab test may access and modify many items of data. Without careful preparation before a test runs, and cleanup afterward, data can be changed by a test in such a way that it affects the behaviour of following tests. This should be avoided at all costs! Fortunately, the existing test framework handles most cases already.

When the test environment does get polluted, a common outcome is flaky tests. Pollution will often manifest as an order dependency: running spec A followed by spec B will reliably fail, but running spec B followed by spec A will reliably succeed. In these cases, you can use rspec --bisect (or a manual pairwise bisect of spec files) to determine which spec is at fault. Fixing the problem requires some understanding of how the test suite ensures the environment is pristine. Read on to discover more about each data store!

SQL database

This is managed for us by the database_cleaner gem. Each spec is surrounded in a transaction, which is rolled back once the test completes. Certain specs will instead issue DELETE FROM queries against every table after completion; this allows the created rows to be viewed from multiple database connections, which is important for specs that run in a browser, or migration specs, among others.

One consequence of using these strategies, instead of the well-known TRUNCATE TABLES approach, is that primary keys and other sequences are not reset across specs. So if you create a project in spec A, then create a project in spec B, the first will have id=1, while the second will have id=2.

This means that specs should never rely on the value of an ID, or any other sequence-generated column. To avoid accidental conflicts, specs should also avoid manually specifying any values in these kinds of columns. Instead, leave them unspecified, and look up the value after the row is created.

Redis

GitLab stores two main categories of data in Redis: cached items, and Sidekiq jobs.

In most specs, the Rails cache is actually an in-memory store. This is replaced between specs, so calls to Rails.cache.read and Rails.cache.write are safe. However, if a spec makes direct Redis calls, it should mark itself with the :clean_gitlab_redis_cache, :clean_gitlab_redis_shared_state or :clean_gitlab_redis_queues traits as appropriate.

Background jobs / Sidekiq

By default, Sidekiq jobs are enqueued into a jobs array and aren’t processed. If a test enqueues Sidekiq jobs and need them to be processed, the :sidekiq_inline trait can be used.

The :sidekiq_might_not_need_inline trait was added when Sidekiq inline mode was changed to fake mode to all the tests that needed Sidekiq to actually process jobs. Tests with this trait should be either fixed to not rely on Sidekiq processing jobs, or their :sidekiq_might_not_need_inline trait should be updated to :sidekiq_inline if the processing of background jobs is needed/expected.

Note: The usage of perform_enqueued_jobs is currently useless since our workers aren’t inheriting from ApplicationJob / ActiveJob::Base.

DNS

DNS requests are stubbed universally in the test suite (as of !22368), as DNS can cause issues depending on the developer’s local network. There are RSpec labels available in spec/support/dns.rb which you can apply to tests if you need to bypass the DNS stubbing, e.g.:

it "really connects to Prometheus", :permit_dns do

And if you need more specific control, the DNS blocking is implemented in spec/support/helpers/dns_helpers.rb and these methods can be called elsewhere.

Filesystem

Filesystem data can be roughly split into “repositories”, and “everything else”. Repositories are stored in tmp/tests/repositories. This directory is emptied before a test run starts, and after the test run ends. It is not emptied between specs, so created repositories accumulate within this directory over the lifetime of the process. Deleting them is expensive, but this could lead to pollution unless carefully managed.

To avoid this, hashed storage is enabled in the test suite. This means that repositories are given a unique path that depends on their project’s ID. Since the project IDs are not reset between specs, this guarantees that each spec gets its own repository on disk, and prevents changes from being visible between specs.

If a spec manually specifies a project ID, or inspects the state of the tmp/tests/repositories/ directory directly, then it should clean up the directory both before and after it runs. In general, these patterns should be completely avoided.

Other classes of file linked to database objects, such as uploads, are generally managed in the same way. With hashed storage enabled in the specs, they are written to disk in locations determined by ID, so conflicts should not occur.

Some specs disable hashed storage by passing the :legacy_storage trait to the projects factory. Specs that do this must never override the path of the project, or any of its groups. The default path includes the project ID, so will not conflict; but if two specs create a :legacy_storage project with the same path, they will use the same repository on disk and lead to test environment pollution.

Other files must be managed manually by the spec. If you run code that creates a tmp/test-file.csv file, for instance, the spec must ensure that the file is removed as part of cleanup.

Persistent in-memory application state

All the specs in a given rspec run share the same Ruby process, which means they can affect each other by modifying Ruby objects that are accessible between specs. In practice, this means global variables, and constants (which includes Ruby classes, modules, etc).

Global variables should generally not be modified. If absolutely necessary, a block like this can be used to ensure the change is rolled back afterwards:

around(:each) do |example|
  old_value = $0

  begin
    $0 = "new-value"
    example.run
  ensure
    $0 = old_value
  end
end

If a spec needs to modify a constant, it should use the stub_const helper to ensure the change is rolled back.

If you need to modify the contents of the ENV constant, you can use the stub_env helper method instead.

While most Ruby instances are not shared between specs, classes and modules generally are. Class and module instance variables, accessors, class variables, and other stateful idioms, should be treated in the same way as global variables - don’t modify them unless you have to! In particular, prefer using expectations, or dependency injection along with stubs, to avoid the need for modifications. If you have no other choice, an around block similar to the example for global variables, above, can be used, but this should be avoided if at all possible.

Table-based / Parameterized tests

This style of testing is used to exercise one piece of code with a comprehensive range of inputs. By specifying the test case once, alongside a table of inputs and the expected output for each, your tests can be made easier to read and more compact.

We use the rspec-parameterized gem. A short example, using the table syntax and checking Ruby equality for a range of inputs, might look like this:

describe "#==" do
  using RSpec::Parameterized::TableSyntax

  where(:a, :b, :result) do
    1         | 1        | true
    1         | 2        | false
    true      | true     | true
    true      | false    | false
  end

  with_them do
    it { expect(a == b).to eq(result) }

    it 'is isomorphic' do
      expect(b == a).to eq(result)
    end
  end
end
Caution: Only use simple values as input in the where block. Using procs, stateful objects, FactoryBot-created objects etc. can lead to unexpected results.

Prometheus tests

Prometheus metrics may be preserved from one test run to another. To ensure that metrics are reset before each example, add the :prometheus tag to the Rspec test.

Matchers

Custom matchers should be created to clarify the intent and/or hide the complexity of RSpec expectations.They should be placed under spec/support/matchers/. Matchers can be placed in subfolder if they apply to a certain type of specs only (e.g. features, requests etc.) but shouldn’t be if they apply to multiple type of specs.

be_like_time

Time returned from a database can differ in precision from time objects in Ruby, so we need flexible tolerances when comparing in specs. We can use be_like_time to compare that times are within one second of each other.

Example:

expect(metrics.merged_at).to be_like_time(time)

have_gitlab_http_status

Prefer have_gitlab_http_status over have_http_status because the former could also show the response body whenever the status mismatched. This would be very useful whenever some tests start breaking and we would love to know why without editing the source and rerun the tests.

This is especially useful whenever it’s showing 500 internal server error.

Prefer named HTTP status like :no_content over its numeric representation 206. See a list of supported status codes.

Example:

expect(response).to have_gitlab_http_status(:ok)

Shared contexts

Shared contexts only used in one spec file can be declared inline. Any shared contexts used by more than one spec file:

  • Should be placed under spec/support/shared_contexts/.
  • Can be placed in subfolder if they apply to a certain type of specs only (e.g. features, requests etc.) but shouldn’t be if they apply to multiple type of specs.

Each file should include only one context and have a descriptive name, e.g. spec/support/shared_contexts/controllers/githubish_import_controller_shared_context.rb.

Shared examples

Shared examples only used in one spec file can be declared inline. Any shared examples used by more than one spec file:

  • Should be placed under spec/support/shared_examples/.
  • Can be placed in subfolder if they apply to a certain type of specs only (e.g. features, requests etc.) but shouldn’t be if they apply to multiple type of specs.

Each file should include only one context and have a descriptive name, e.g. spec/support/shared_examples/controllers/githubish_import_controller_shared_example.rb.

Helpers

Helpers are usually modules that provide some methods to hide the complexity of specific RSpec examples. You can define helpers in RSpec files if they’re not intended to be shared with other specs. Otherwise, they should be placed under spec/support/helpers/. Helpers can be placed in subfolder if they apply to a certain type of specs only (e.g. features, requests etc.) but shouldn’t be if they apply to multiple type of specs.

Helpers should follow the Rails naming / namespacing convention. For instance spec/support/helpers/cycle_analytics_helpers.rb should define:

module Spec
  module Support
    module Helpers
      module CycleAnalyticsHelpers
        def create_commit_referencing_issue(issue, branch_name: random_git_name)
          project.repository.add_branch(user, branch_name, 'master')
          create_commit("Commit for ##{issue.iid}", issue.project, user, branch_name)
        end
      end
    end
  end
end

Helpers should not change the RSpec config. For instance, the helpers module described above should not include:

RSpec.configure do |config|
  config.include Spec::Support::Helpers::CycleAnalyticsHelpers
end

Factories

GitLab uses factory_bot as a test fixture replacement.

  • Factory definitions live in spec/factories/, named using the pluralization of their corresponding model (User factories are defined in users.rb).
  • There should be only one top-level factory definition per file.
  • FactoryBot methods are mixed in to all RSpec groups. This means you can (and should) call create(...) instead of FactoryBot.create(...).
  • Make use of traits to clean up definitions and usages.
  • When defining a factory, don’t define attributes that are not required for the resulting record to pass validation.
  • When instantiating from a factory, don’t supply attributes that aren’t required by the test.
  • Factories don’t have to be limited to ActiveRecord objects. See example.

Fixtures

All fixtures should be placed under spec/fixtures/.

Repositories

Testing some functionality, e.g., merging a merge request, requires a Git repository with a certain state to be present in the test environment. GitLab maintains the gitlab-test repository for certain common cases - you can ensure a copy of the repository is used with the :repository trait for project factories:

let(:project) { create(:project, :repository) }

Where you can, consider using the :custom_repo trait instead of :repository. This allows you to specify exactly what files will appear in the master branch of the project’s repository. For example:

let(:project) do
  create(
    :project, :custom_repo,
    files: {
      'README.md'       => 'Content here',
      'foo/bar/baz.txt' => 'More content here'
    }
  )
end

This will create a repository containing two files, with default permissions and the specified content.

Config

RSpec config files are files that change the RSpec config (i.e. RSpec.configure do |config| blocks). They should be placed under spec/support/.

Each file should be related to a specific domain, e.g. spec/support/capybara.rb, spec/support/carrierwave.rb, etc.

If a helpers module applies only to a certain kind of specs, it should add modifiers to the config.include call. For instance if spec/support/helpers/cycle_analytics_helpers.rb applies to :lib and type: :model specs only, you would write the following:

RSpec.configure do |config|
  config.include Spec::Support::Helpers::CycleAnalyticsHelpers, :lib
  config.include Spec::Support::Helpers::CycleAnalyticsHelpers, type: :model
end

If a config file only consists of config.include, you can add these config.include directly in spec/spec_helper.rb.

For very generic helpers, consider including them in the spec/support/rspec.rb file which is used by the spec/fast_spec_helper.rb file. See Fast unit tests for more details about the spec/fast_spec_helper.rb file.

Return to Testing documentation