parallel_tests

Speedup Test::Unit + RSpec + Cucumber + Spinach by running parallel on multiple CPU cores.
ParallelTests splits tests into even groups (by number of lines or runtime) and runs each group in a single process with its own database.

Setup for Rails

RailsCasts episode #413 Fast Tests

Install

Gemfile:

gem 'parallel_tests', group: [:development, :test]

Add to config/database.yml

ParallelTests uses 1 database per test-process.

Process number	1	2	3
ENV['TEST_ENV_NUMBER']	''	'2'	'3'

test:
  database: yourproject_test<%= ENV['TEST_ENV_NUMBER'] %>

Create additional database(s)

rake parallel:create

Copy development schema (repeat after migrations)

rake parallel:prepare

Setup environment from scratch (create db and loads schema, useful for CI)

rake parallel:setup

Run!

rake parallel:test          # Test::Unit
rake parallel:spec          # RSpec
rake parallel:features      # Cucumber
rake parallel:features-spinach       # Spinach

rake parallel:test[1] --> force 1 CPU --> 86 seconds
rake parallel:test    --> got 2 CPUs? --> 47 seconds
rake parallel:test    --> got 4 CPUs? --> 26 seconds
...

Test by pattern (e.g. use one integration server per subfolder / see if you broke any 'user'-related tests)

rake parallel:test[^test/unit] # every test file in test/unit folder
rake parallel:test[user]  # run users_controller + user_helper + user tests
rake parallel:test['user|product']  # run user and product related tests

Example output

2 processes for 210 specs, ~ 105 specs per process
... test output ...

843 examples, 0 failures, 1 pending

Took 29.925333 seconds

Run an arbitrary task in parallel

RAILS_ENV=test parallel_test -e "rake my:custom:task"
# or
rake parallel:rake[my:custom:task]
# limited parallelism
rake parallel:rake[my:custom:task,2]

Running things once

# preparation:
# affected by race-condition: first process may boot slower than the second
# either sleep a bit or use a lock for example File.lock
ParallelTests.first_process? ? do_something : sleep(1)

# cleanup: 
# last_process? does NOT mean last finished process, just last started
ParallelTests.last_process? ? do_something : sleep(1)

at_exit do
  if ParallelTests.first_process?
    ParallelTests.wait_for_other_processes_to_finish
    undo_something
  end
end

Loggers

Even test group run-times

RSpec

Add the RuntimeLogger to log how long each test takes to run. This log file will be loaded on the next test run, and the tests will be grouped so that each process should finish around the same time.

Rspec: Add to your .rspec_parallel (or .rspec) :

--format progress
--format ParallelTests::RSpec::RuntimeLogger --out tmp/parallel_runtime_rspec.log

Test::Unit & Minitest 4/5

Add to your test_helper.rb:

require 'parallel_tests/test/runtime_logger' if ENV['RECORD_RUNTIME']

results will be logged to tmp/parallel_runtime_test.log when RECORD_RUNTIME is set, so it is not always required or overwritten.

RSpec: SummaryLogger

Log the test output without the different processes overwriting each other.

Add the following to your .rspec_parallel (or .rspec) :

--format progress
--format ParallelTests::RSpec::SummaryLogger --out tmp/spec_summary.log

RSpec: FailuresLogger

Produce pasteable command-line snippets for each failed example.

E.g.

rspec /path/to/my_spec.rb:123 # should do something

Add the following to your .rspec_parallel (or .rspec) :

--format progress
--format ParallelTests::RSpec::FailuresLogger --out tmp/failing_specs.log

Cucumber: FailuresLogger

Log failed cucumber scenarios to the specified file. The filename can be passed to cucumber, prefixed with '@' to rerun failures.

Usage:

cucumber --format ParallelTests::Cucumber::FailuresLogger --out tmp/cucumber_failures.log

Or add the formatter to the parallel: profile of your cucumber.yml:

parallel: --format progress --format ParallelTests::Cucumber::FailuresLogger --out tmp/cucumber_failures.log

Note if your cucumber.yml default profile uses <%= std_opts %> you may need to insert this as follows parallel: <%= std_opts %> --format progress...

To rerun failures:

cucumber @tmp/cucumber_failures.log

Setup for non-rails

gem install parallel_tests
# go to your project dir
parallel_test test/
parallel_rspec spec/
parallel_cucumber features/
parallel_spinach features/

• use ENV['TEST_ENV_NUMBER'] inside your tests to select separate db/memcache/etc.

• Only run selected files & folders:

  parallel_test test/bar test/baz/foo_text.rb

• Pass test-options and files via --:

  parallel_test -- -t acceptance -f progress -- spec/foo_spec.rb spec/acceptance

Options are:

-n [PROCESSES]                   How many processes to use, default: available CPUs
-p, --pattern [PATTERN]          run tests matching this regex pattern
    --group-by [TYPE]            group tests by:
      found - order of finding files
      steps - number of cucumber/spinach steps
      scenarios - individual cucumber scenarios
      filesize - by size of the file
      runtime - info from runtime log
      default - runtime when runtime log is filled otherwise filesize
-m, --multiply-processes [FLOAT] use given number as a multiplier of processes to run
-s, --single [PATTERN]           Run all matching files in the same process
-i, --isolate                    Do not run any other tests in the group used by --single(-s)
    --only-group INT[, INT]
-e, --exec [COMMAND]             execute this code parallel and with ENV['TEST_ENV_NUMBER']
-o, --test-options '[OPTIONS]'   execute test commands with those options
-t, --type [TYPE]                test(default) / rspec / cucumber / spinach
    --suffix [PATTERN]           override built in test file pattern (should match suffix):
      '_spec.rb$' - matches rspec files
      '_(test|spec).rb$' - matches test or spec files
    --serialize-stdout           Serialize stdout output, nothing will be written until everything is done
    --combine-stderr             Combine stderr into stdout, useful in conjunction with --serialize-stdout
    --non-parallel               execute same commands but do not in parallel, needs --exec
    --no-symlinks                Do not traverse symbolic links to find test files
    --ignore-tags [PATTERN]      When counting steps ignore scenarios with tags that match this pattern
    --nice                       execute test commands with low priority.
    --runtime-log [PATH]         Location of previously recorded test runtimes
    --allowed-missing            Allowed percentage of missing runtimes (default = 50)
    --unknown-runtime [FLOAT]    Use given number as unknown runtime (otherwise use average time)
    --verbose                    Print more output
-v, --version                    Show Version
-h, --help                       Show this.

You can run any kind of code in parallel with -e / --exec

parallel_test -n 5 -e 'ruby -e "puts %[hello from process #{ENV[:TEST_ENV_NUMBER.to_s].inspect}]"'
hello from process "2"
hello from process ""
hello from process "3"
hello from process "5"
hello from process "4"

	1 Process	2 Processes	4 Processes
RSpec spec-suite	18s	14s	10s
Rails-ActionPack	88s	53s	44s

TIPS

RSpec

• Add a .rspec_parallel to use different options, e.g. no --drb
• Remove --loadby from .rspec
• Instantly see failures (instead of just a red F) with rspec-instafail
• Use rspec-retry (not rspec-rerun) to rerun failed tests.
• JUnit formatter configuration

Cucumber

• Add a parallel: foo profile to your config/cucumber.yml and it will be used to run parallel tests
• ReportBuilder can help with combining parallel test results
  • Supports Cucumber 2.0+ and is actively maintained
  • Combines many JSON files into a single file
  • Builds a HTML report from JSON with support for debug msgs & embedded Base64 images.

General

• [SQL schema format] use :ruby schema format to get faster parallel:prepare`
• [ZSH] use quotes to use rake arguments rake "parallel:prepare[3]"
• [Memcached] use different namespaces
  e.g. config.cache_store = ..., namespace: "test_#{ENV['TEST_ENV_NUMBER']}"
• Debug errors that only happen with multiple files using --verbose and cleanser
• export PARALLEL_TEST_PROCESSORS=13 to override default processor count
• Shell alias: alias prspec='parallel_rspec -m 2 --'
• [Spring] to use spring you have to patch it
• --first-is-1 will make the first environment be 1, so you can test while running your full suite.
  export PARALLEL_TEST_FIRST_IS_1=true will provide the same result
• email_spec and/or action_mailer_cache_delivery
• zeus-parallel_tests
• Distributed parallel test (e.g. Travis Support)
• Capybara setup
• Sphinx setup
• Capistrano setup let your tests run on a big box instead of your laptop

Contribute your own gotaches to the Wiki or even better open a PR :)

TODO

• fix tests vs cucumber >= 1.2 unknown option --format
• add unit tests for cucumber runtime formatter
• make windows compatible

Authors

inspired by pivotal labs

Contributors

• Charles Finkel
• Indrek Juhkam
• Jason Morrison
• jinzhu
• Joakim Kolsjö
• Kevin Scaldeferri
• Kpumuk
• Maksim Horbul
• Pivotal Labs
• Rohan Deshpande
• Tchandy
• Terence Lee
• Will Bryant
• Fred Wu
• xxx
• Levent Ali
• Michael Kintzer
• nathansobo
• Joe Yates
• asmega
• Doug Barth
• Geoffrey Hichborn
• Trae Robrock
• Lawrence Wang
• Sean Walbran
• Lawrence Wang
• Potapov Sergey
• Łukasz Tackowiak
• Pedro Carriço
• Pablo Manrubia Díez
• Slawomir Smiechura
• Georg Friedrich
• R. Tyler Croy
• Ulrich Berkmüller
• Grzegorz Derebecki
• Florian Motlik
• Artem Kuzko
• Zeke Fast
• Joseph Shraibman
• David Davis
• Ari Pollak
• Aaron Jensen
• Artur Roszczyk
• Caleb Tomlinson
• Jawwad Ahmad
• Iain Beeston
• Alejandro Pulver
• Felix Clack
• Izaak Alpert
• Micah Geisel
• Exoth
• sidfarkus
• Colin Harris
• Wataru MIYAGUNI
• Brandon Turner
• Matt Hodgson
• bicarbon8
• seichner
• Matt Southerden
• Stanislaw Wozniak
• Dmitry Polushkin
• Samer Masry
• Volodymyr Mykhailyk
• Mike Mueller
• Aaron Jensen
• Ed Slocomb
• Cezary Baginski
• Marius Ioana
• Lukas Oberhuber
• Ryan Zhang
• Rhett Sutphin
• Doc Ritezel

Michael Grosser
michael@grosser.it
License: MIT