Ruby Can Too Scale: Highly Performant Microservices in Ruby (OSCON 2016)

Ruby developers have known for years that our beloved language is one of the most joyful to use. And yet, “Ruby Can’t Scale” seems to be a daily article on Hacker News. Tim Krajcar presents some tips and tricks that he’s learned at New Relic writing Ruby services that scale to hundreds of thousands of requests per minute.

Topics include:

* Building performance into your architecture from your first commit
* Where to invest precious development hours to realize the most performance value
* Which frameworks to use, and which to avoid, when working at scale
* Designing to avoid cascading failures when building highly interdependent services

Transcript

1. Ruby Can Too Scale:
   Highly Performant Microservices in Ruby
   Tim Krajcar
   

   View Slide

2. April 2015
   

   View Slide

3. TimKrajcar
   \
   

   View Slide

4. TimKrajcar
   This document and the information herein (including any information that may be incorporated by reference) is provided for
   informational purposes only and should not be construed as an offer, commitment, promise or obligation on behalf of New Relic, Inc.
   (“New Relic”) to sell securities or deliver any product, material, code, functionality, or other feature. Any information provided hereby
   is proprietary to New Relic and may not be replicated or disclosed without New Relic’s express written permission.
   Such information may contain forward-looking statements within the meaning of federal securities laws. Any statement that is not a
   historical fact or refers to expectations, projections, future plans, objectives, estimates, goals, or other characterizations of future
   events is a forward-looking statement. These forward-looking statements can often be identified as such because the context of the
   statement will include words such as “believes,” “anticipates,” “expects” or words of similar import.
   Actual results may differ materially from those expressed in these forward-looking statements, which speak only as of the date
   hereof, and are subject to change at any time without notice. Existing and prospective investors, customers and other third parties
   transacting business with New Relic are cautioned not to place undue reliance on this forward-looking information. The achievement
   or success of the matters covered by such forward-looking statements are based on New Relic’s current assumptions, expectations,
   and beliefs and are subject to substantial risks, uncertainties, assumptions, and changes in circumstances that may cause the actual
   results, performance, or achievements to differ materially from those expressed or implied in any forward-looking statement. Further
   information on factors that could affect such forward-looking statements is included in the filings we make with the SEC from time
   to time. Copies of these documents may be obtained by visiting New Relic’s Investor Relations website at ir.newrelic.com or the
   SEC’s website at www.sec.gov.
   New Relic assumes no obligation and does not intend to update these forward-looking statements, except as required by law. New
   Relic makes no warranties, expressed or implied, in this document or otherwise, with respect to the information provided.
   

   View Slide

5. TimKrajcar
   

   View Slide

6. TimKrajcar
   why this talk?
   

   View Slide

7. View Slide

8. View Slide

9. View Slide

10. View Slide

11. TimKrajcar
    "ruby can't scale"
    

    View Slide

12. –Johnny Appleseed
    

    View Slide

13. –Johnny Appleseed
    

    View Slide

14. –Johnny Appleseed
    

    View Slide

15. TimKrajcar
    "ruby can't scale"
    

    View Slide

16. TimKrajcar
    "rails can't scale"
    

    View Slide

17. TimKrajcar
    second system
    effect
    

    View Slide

18. –Johnny Appleseed
    

    View Slide

19. –Johnny Appleseed
    and, even more importantly, stack
    

    View Slide

20. TimKrajcar
    why ruby?
    

    View Slide

21. TimKrajcar
    expressive programming language
    =
    fun to write code
    =
    I want to write more of it
    

    View Slide

22. TimKrajcar
    expressive programming language
    =
    fun to write code
    =
    I want to write more of it
    productivity!!
    
    

    View Slide

23. TimKrajcar
    things not covered
    

    View Slide

24. TimKrajcar
    1. database tuning
    

    View Slide

25. TimKrajcar
    1. database tuning
    2. caching
    

    View Slide

26. TimKrajcar
    1. database tuning
    2. caching
    3. network optimization
    

    View Slide

27. TimKrajcar
    1. database tuning
    2. caching
    3. network optimization
    4. browser-y things
    

    View Slide

28. TimKrajcar
    microservices at New
    Relic
    

    View Slide

29. TimKrajcar
    ~60 total services
    

    View Slide

30. TimKrajcar
    some not so micro
    

    View Slide

31. TimKrajcar
    Ruby:
    common for prototypes
    

    View Slide

32. TimKrajcar
    templates and auto-
    generation
    

    View Slide

33. –Johnny Appleseed
    

    View Slide

34. TimKrajcar
    stereotypical example:
    customer permissions service
    

    View Slide

35. –Johnny Appleseed
    

    View Slide

36. –Johnny Appleseed
    

    View Slide

37. PERFORMANCE FROM THE
    FIRST COMMIT
    

    View Slide

38. –Johnny Appleseed
    Structured programming with 'go to' statements
    Donald E. Knuth
    COMPUTING SURVEYS, DECEMBER 1974
    

    View Slide

39. –Johnny Appleseed
    We should forget about small
    efficiencies, say about 97% of the
    time: premature optimization is the
    root of all evil.
    

    View Slide

40. –Johnny Appleseed
    Yet we should not pass up our opportunities in
    that critical 3%. A good programmer will not be
    lulled into complacency by such reasoning,
    [and] will be wise to look carefully at the critical
    code.
    

    View Slide

41. TimKrajcar
    benchmarking
    methodologies matter
    

    View Slide

42. TimKrajcar
    app server time
    vs
    request queuing
    

    View Slide

43. –Johnny Appleseed
    

    View Slide

44. –Johnny Appleseed
    

    View Slide

45. TimKrajcar
    X-Request-Start
    HTTP header
    

    View Slide

46. TimKrajcar
    test suite tools
    

    View Slide

47. TimKrajcar
    minitest-perf
    https://github.com/mrsimo/minitest-perf
    

    View Slide

48. > bundle exec minitest-perf
    Slowest individual tests
    1210.00ms | NewRelic::Agent::AgentTest#test_after_fork_should_
    1190.78ms | LicenseTest#test_for_scary_license_terms
    1034.58ms | NewRelic::Agent::Agent::ConnectTest#test_connect_g
    1034.20ms | NewRelic::Agent::Agent::ConnectTest#test_logging_c
    1032.60ms | NewRelic::Agent::Agent::ConnectTest#test_environme
    1031.61ms | NewRelic::Agent::Agent::ConnectTest#test_connect_s
    1030.49ms | NewRelic::Agent::Agent::ConnectTest#test_connect_s
    730.88ms | OrphanedConfigTest#test_all_default_source_config_
    455.05ms | NewRelic::Agent::MetricStatsTest#test_record_scope
    434.18ms | NewRelicServiceTest#test_json_marshaller_should_ha
    

    View Slide

49. Slowest test suites
    638.87ms | 2 | LicenseTest
    293.86ms | 3 | OrphanedConfigTest
    193.41ms | 27 | NewRelic::Agent::Agent::ConnectTest
    191.36ms | 1 | NewRelic::Agent::Threading::BacktraceServic
    130.64ms | 14 | NewRelic::Agent::PipeChannelManagerTest
    118.76ms | 14 | PipeServiceTest
    111.14ms | 4 | DispatcherTest
    32.82ms | 63 | NewRelic::Agent::AgentTest
    25.50ms | 20 | NewRelic::Agent::MetricStatsTest
    19.37ms | 12 | NewRelic::Agent::UtilizationDataTest
    

    View Slide

50. TimKrajcar
    framework choices
    

    View Slide

51. TimKrajcar
    framework choices
    reductions
    

    View Slide

52. –Johnny Appleseed
    

    View Slide

53. TimKrajcar
    example!!
    

    View Slide

54. TimKrajcar
    a typical Ruby microservice:
    REST DB-to-JSON API
    https://github.com/tkrajcar/ruby-can-too-scale
    

    View Slide

55. TimKrajcar
    a typical Ruby microservice:
    REST DB-to-JSON API
    https://github.com/tkrajcar/ruby-can-too-scale
    !! warning: YOLOCODE !!
    

    View Slide

56. TimKrajcar
    •Rails 4.2.6 straight outta rails new + Puma +
    nginx
    •JSON only, using scaffolded controller & model
    •testing #index (~25kb) and #show (<1 kb)
    •read-only
    •Dedicated EC2 c3.large instances (2 vCPUs)
    •Postgres on its own dedicated RDS instance
    •30 users as fast as possible (separate instance)
    

    View Slide

57. TimKrajcar
    best metric:
    client requests per minute
    

    View Slide

58. TimKrajcar
    #index
    /articles.json
    #show
    /articles/1.json
    Stock Rails scaffolding 4,211 18,011
    

    View Slide

59. TimKrajcar
    #index
    /articles.json
    #show
    /articles/1.json
    Stock Rails scaffolding 4,211 18,011
    No jbuilder templates
    4,561
    +8.3%
    18,909
    +4.9%
    

    View Slide

60. TimKrajcar
    #index
    /articles.json
    #show
    /articles/1.json
    Stock Rails scaffolding 4,211 18,011
    No jbuilder templates
    4,561
    +8.3%
    18,909
    +4.9%
    oj gem
    5,548
    +21.6%
    19,714
    +9.4%
    

    View Slide

61. TimKrajcar
    #index
    /articles.json
    #show
    /articles/1.json
    Stock Rails scaffolding 4,211 18,011
    No jbuilder templates
    4,561
    +8.3%
    18,909
    +4.9%
    oj gem
    5,548
    +21.6%
    19,714
    +9.4%
    Sequel instead of
    ActiveRecord
    7,035
    +26.8%
    22,378
    +13.5%
    

    View Slide

62. TimKrajcar
    #index
    /articles.json
    #show
    /articles/1.json
    Stock Rails scaffolding 4,211 18,011
    No jbuilder templates
    4,561
    +8.3%
    18,909
    +4.9%
    oj gem
    5,548
    +21.6%
    19,714
    +9.4%
    Sequel instead of
    ActiveRecord
    7,035
    +26.8%
    22,378
    +13.5%
    Sinatra
    (with Sequel and oj)
    13,245 85,017
    

    View Slide

63. TimKrajcar
    #index
    /articles.json
    #show
    /articles/1.json
    Stock Rails scaffolding 4,211 18,011
    No jbuilder templates
    4,561
    +8.3%
    18,909
    +4.9%
    oj gem
    5,548
    +21.6%
    19,714
    +9.4%
    Sequel instead of
    ActiveRecord
    7,035
    +26.8%
    22,378
    +13.5%
    Sinatra
    (with Sequel and oj)
    13,245 85,017
    
    
    

    View Slide

64. TimKrajcar
    tl;dr:
    use Sinatra and Sequel
    

    View Slide

65. DIVING DEEPER
    

    View Slide

66. TimKrajcar
    Derailed Benchmarks
    https://github.com/schneems/derailed_benchmarks
    

    View Slide

67. > bundle exec derailed bundle:mem
    TOP: 46.5625 MiB
    activeadmin: 17.5156 MiB
    active_admin: 17.5039 MiB
    meta_search: 6.2773 MiB
    meta_search/searches/active_record: 1.707 MiB
    meta_search/builder: 1.5938 MiB
    meta_search/where: 1.0977 MiB
    polyamorous: 0.4688 MiB
    active_admin/comments: 3.6055 MiB
    active_admin/comments/views: 1.375 MiB
    active_admin/views: 1.2305 MiB (Also required by:
    active_admin/comments/views/active_admin_comments)
    /Users/tkrajcar/.rbenv/versions/2.1.7/lib/ruby/gems/
    2.1.0/gems/activeadmin-0.6.3/lib/active_admin/views/components/
    paginated_collection.rb: 0.3125 MiB
    bourbon: 3.3438 MiB
    

    View Slide

68. > bundle exec derailed bundle:objects
    Measuring objects created by gems in groups [:default, "production"]
    Total allocated: 72414462 bytes (425465 objects)
    Total retained: 24736698 bytes (44790 objects)
    allocated memory by gem
    -----------------------------------
    27407703 activesupport-3.2.22.2
    9240545 erubis-2.7.0
    5722758 actionpack-3.2.22.2
    3287857 activerecord-3.2.22.2
    3176375 haml-4.0.5
    1728840 meta_search-1.1.3
    1300638 devise-1.5.4
    1216006 2.1.7/lib
    1016914 formtastic-2.2.1
    993225 activeadmin-0.6.3
    

    View Slide

69. TimKrajcar
    Stackprof
    https://github.com/tmm1/stackprof
    

    View Slide

70. –Johnny Appleseed
    

    View Slide

71. TimKrajcar
    other interpreters
    

    View Slide

72. DESIGNING TO AVOID
    CASCADING FAILURES
    

    View Slide

73. TimKrajcar
    timeouts
    

    View Slide

74. –Johnny Appleseed
    

    View Slide

75. View Slide

76. TimKrajcar
    Timeout
    

    View Slide

77. require 'timeout'
    result = begin
    Timeout::timeout(0.5) { do_a_possibly_slow_thing }
    rescue Timeout::Error
    # log/notify about the timeout
    0
    end
    

    View Slide

78. View Slide

79. View Slide

80. View Slide

81. TimKrajcar
    circuit breaker pattern
    

    View Slide

82. –Johnny Appleseed
    http://martinfowler.com/bliki/CircuitBreaker.html
    

    View Slide

83. TimKrajcar
    cb2
    https://github.com/pedro/cb2
    

    View Slide

84. require 'cb2'
    breaker = CB2::Breaker.new(
    service: "cashwings", # identify each circuit breaker individually
    duration: 60, # in this amount of seconds...
    threshold: 5, # ...this % of errors will open the breaker
    reenable_after: 600, # once opened, how long until we try again
    redis: Redis.new) # uses Redis to keep track by service name
    

    View Slide

85. require 'cb2'
    breaker = CB2::Breaker.new(
    service: "cashwings", # identify each circuit breaker individually
    duration: 60, # in this amount of seconds...
    threshold: 5, # ...this % of errors will open the breaker
    reenable_after: 600, # once opened, how long until we try again
    redis: Redis.new) # uses Redis to keep track by service name
    
    

    View Slide

86. begin
    breaker.run do
    some_api_request()
    end
    rescue CB2::BreakerOpen
    # log/notify about the breaker opening!
    alternate_response()
    end
    

    View Slide

87. TimKrajcar
    parallelize
    

    View Slide

88. TimKrajcar
    parallelize everything
    especially HTTP
    

    View Slide

89. TimKrajcar
    Typhoeus::Hydra
    https://github.com/typhoeus/typhoeus
    

    View Slide

90. require 'typhoeus'
    request1 = Typhoeus::Request.new("http://example.com/1.json")
    request2 = Typhoeus::Request.new("http://example.com/2.json")
    hydra = Typhoeus::Hydra.new
    hydra.queue(request1)
    hydra.queue(request2)
    hydra.run # blocks here until queue finishes
    puts "Request 1 status #{request1.response.code}"
    puts "Request 2 status #{request2.response.code}"
    

    View Slide

91. TimKrajcar
    background everything*
    *unless you can't
    

    View Slide

92. TimKrajcar
    https://github.com/mperham/sidekiq
    

    View Slide

93. # app/workers/welcome_emailer.rb
    class WelcomeEmailer
    include Sidekiq::Worker
    def perform(email)
    send_email_to(email, "Welcome!")
    end
    end
    # app/controllers/welcome_controller.rb
    def welcome
    # blah blah create a @user
    WelcomeEmailer.perform_async(@user.email)
    end
    

    View Slide

94. # app/workers/welcome_emailer.rb
    class WelcomeEmailer
    include Sidekiq::Worker
    def perform(email)
    send_email_to(email, "Welcome!")
    end
    end
    # app/controllers/welcome_controller.rb
    def welcome
    # blah blah create a @user
    WelcomeEmailer.perform_async(@user.email)
    end
    doesn't
    block
    

    View Slide

95. TimKrajcar
    rate limit by client
    

    View Slide

96. TimKrajcar
    Rack::Throttle
    https://github.com/bendiken/rack-throttle
    

    View Slide

97. # config/application.rb
    require 'rack/throttle'
    class Application < Rails::Application
    config.middleware.use Rack::Throttle::Hourly, max: 100
    end
    

    View Slide

98. # config/application.rb
    require 'rack/throttle'
    class Application < Rails::Application
    config.middleware.use Rack::Throttle::Hourly, max: 100
    end
    client = by IP
    or you can define custom client IDs
    

    View Slide

99. TimKrajcar
    monitoring
    &
    alerting
    

    View Slide

100. TimKrajcar
     low traffic?
     drive synthetic traffic
     

     View Slide

101. TimKrajcar
     synthetic-monitor
     https://github.com/johnboyes/synthetic-monitor
     

     View Slide

102. View Slide

103. TimKrajcar
     RECAP!!
     

     View Slide

104. TimKrajcar
     1. use Ruby to write services, because it's fun!
     2. monitor both request queuing time and app server time
     3. make informed framework decisions (use sinatra)
     4. set real timeouts and use circuit breakers
     5. parallelize everything
     6. background everything
     7. rate limit by client, even for internal clients
     8. drive synthetic traffic to help catch problems right away
     

     View Slide

105. TimKrajcar
     references
     

     View Slide

106. TimKrajcar
     Nate Berkopec: nateberkopec.com
     High Scalability: highscalability.com
     Improved Production Stability With Circuit
     Breakers: blog.heroku.com
     

     View Slide

107. TimKrajcar
     thanks!
     

     View Slide