parallel_tests

parallel_tests

Speedup Minitest + RSpec + Turnip + Cucumber + Spinach by running parallel on multiple CPU cores.
ParallelTests splits tests into balanced groups (by number of lines or runtime) and runs each group in a 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

(Multi-DB) Create individual database

rake parallel:create:<database>
rake parallel:create:secondary

Copy development schema (repeat after migrations)

rake parallel:prepare

Run migrations in additional database(s) (repeat after migrations)

rake parallel:migrate

(Multi-DB) Run migrations in individual database

rake parallel:migrate:<database>

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

rake parallel:setup

Drop all test databases

rake parallel:drop

(Multi-DB) Drop individual test database

rake parallel:drop:<database>

Run!

rake parallel:test          # Minitest
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 with Regex (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
rake "parallel:spec['spec\/(?!features)']" # run RSpec tests except the tests in spec/features

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 setup or teardown once

require "parallel_tests"

# 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

Even test group runtimes

Test groups will often run for different times, making the full test run as slow as the slowest group.

Step 1: Use these loggers (see below) to record test runtime

Step 2: The next test run will use the recorded test runtimes (use --runtime-log <file> if you picked a location different from below)

Step 3: Automate upload/download of test runtime from your CI system example (chunks need to be combined, an alternative is amend)

RSpec

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

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

To use a custom logfile location (default: tmp/parallel_runtime_rspec.log), use the CLI: parallel_test spec -t rspec --runtime-log my.log

Minitest

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.

Loggers

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. For example:

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

Add to .rspec_parallel or use as CLI flag:

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

(Not needed to retry failures, for that pass --only-failures to rspec)

RSpec: VerboseLogger

Prints a single line for starting and finishing each example, to see what is currently running in each process.

# PID, parallel process number, spec status, example description
[14403] [2] [STARTED] Foo foo
[14402] [1] [STARTED] Bar bar
[14402] [1] [PASSED] Bar bar

Add to .rspec_parallel or use as CLI flag:

  --format ParallelTests::RSpec::VerboseLogger

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
parallel_rspec
parallel_cucumber
parallel_spinach

• use ENV['TEST_ENV_NUMBER'] inside your tests to select separate db/memcache/etc. (docker compose: expose it)

• Only run a subset of files / folders:

  parallel_test test/bar test/baz/foo_text.rb

• Pass test-options and files via --:

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

• Pass in test options, by using the -o flag (wrap everything in quotes):

  parallel_cucumber -n 2 -o '-p foo_profile --tags @only_this_tag or @only_that_tag --format summary'

Options are:

-n [PROCESSES]                   How many processes to use, default: available CPUs
-p, --pattern [PATTERN]          run tests matching this regex pattern
    --exclude-pattern [PATTERN]  exclude 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)
    --isolate-n [PROCESSES]      Use 'isolate'  singles with number of processes, default: 1
    --highest-exit-status        Exit with the highest exit status provided by test run(s)
    --failure-exit-code [INT]    Specify the exit code to use when tests fail
    --specify-groups [SPECS]     Use 'specify-groups' if you want to specify multiple specs running in multiple
                                 processes in a specific formation. Commas indicate specs in the same process,
                                 pipes indicate specs in a new process. Cannot use with --single, --isolate, or
                                 --isolate-n.  Ex.
                                 $ parallel_tests -n 3 . --specify-groups '1_spec.rb,2_spec.rb|3_spec.rb'
                                   Process 1 will contain 1_spec.rb and 2_spec.rb
                                   Process 2 will contain 3_spec.rb
                                   Process 3 will contain all other specs
    --only-group INT[,INT]       Only run the given group numbers.
                                 Changes `--group-by` default to 'filesize'.
-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
    --prefix-output-with-test-env-number
                                 Prefixes test env number to the output when not using --serialize-stdout
    --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 [INT]      Allowed percentage of missing runtimes (default = 50)
    --allow-duplicates           When detecting files to run, allow duplicates
    --unknown-runtime [FLOAT]    Use given number as unknown runtime (otherwise use average time)
    --first-is-1                 Use "1" as TEST_ENV_NUMBER to not reuse the default test environment
    --fail-fast                  Stop all groups when one group fails (best used with --test-options '--fail-fast' if supported
    --verbose                    Print debug output
    --verbose-command            Combines options --verbose-process-command and --verbose-rerun-command
    --verbose-process-command    Print the command that will be executed by each process before it begins
    --verbose-rerun-command      After a process fails, print the command executed by that process
    --quiet                      Print only tests 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
• Use parallel_split_test to run multiple scenarios in a single spec file, concurrently. (parallel_tests works at the file-level and intends to stay that way)

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

• [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] Add the spring-commands-parallel-tests gem to your Gemfile to get parallel_tests working with Spring.
• --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 Tests on CI systems) learn how parallel_tests can run on distributed servers such as Travis and GitLab-CI. Also shows you how to use parallel_tests without adding TEST_ENV_NUMBER-backends
• Capybara setup
• Sphinx setup
• Capistrano setup let your tests run on a big box instead of your laptop

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

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
• Alexandre Wilhelm
• Jerry
• Aleksei Gusev
• Scott Olsen
• Andrei Botalov
• Zachary Attas
• David Rodríguez
• Justin Doody
• Sandeep Singh
• Calaway
• alboyadjian
• Nathan Broadbent
• Vikram B Kumar
• Joshua Pinter
• Zach Dennis
• Jon Dufresne
• Eric Kessler
• Adis Osmonov
• Josh Westbrook
• Jay Dorsey
• hatsu

Michael Grosser
[e-mail]
License: MIT