RailsConf 2015 Keynote

Transcript

3. Click to add title Click to add subtitle
4. Why Windows?
5. Kent Beck!!!
6. W IN DO W S
7. No emoji
8. !"#$
9. Welcome To Rails Conf!!!
10. Aaron Patterson
11. @tenderlove
13. Thanks!!!
16. Ruby Central
17. @eileencodes
18. @_ko1
19. Thank You!!! <3
20. Cat photos
24. Extreme Programming
26. Last Year: Thought Leader in Training
27. I Graduated!!
28. Degree in Thought Leadering from Online University
29. Ruby
30. Rails
31. Technoligy
32. Rails
33. Computers
34. Rails
35. Synergy
36. Rails
37. Innovation
38. Rails
39. Future
40. Rails
41. Words
42. Trolling DHH
43. "TDD is dead"
44. SOA is dead
45. Turbo Links 3: RJS
46. I should announce something.
47. Backpack
48. Lightweight
49. Action Fanny Pack
50. Pink
51. Performance! vroom!!!
52. Rails 5?
53. Adequate Development
54. Automatic Parallel Testing
55. Data Base Test Process Test Process Test Process
56. Data Base Test Process Test Process Test Process Data Base Data Base Data Base
57. Automatic
58. Transparent
59. Caching Compiled Views
60. Today: lazy compile
61. Get lock Template Compiled? Compile Save Unlock Writes to memory
62. Threaded Servers: lock
63. Forking servers: waste memory
64. Challenges
65. Too many templates!
66. Rendering with Locals
67. Template source eval(" local_variable = variables_hash[:local_variable] # rest of the template ")
68. Integration Tests
69. Controller Test class ControllerTest < ActionController::TestCase test "index" do get :index assert_equal 200, response.status end end
70. Integration Test class IntegrationTest < ActionDispatch::IntegrationTest test "index" do get '/documents' assert_equal 200, response.status end end
71. Why do we write controller tests?
72. Because integration tests are slow?
73. If tests are slow, is the website slow?
74. Rails 5?: Fast Tests
75. Controller / Integration: What’s the difference?
76. Controller: no middleware
77. Integration: has middleware
78. Rails 5?: Delete controller tests *We won’t actually delete them
79. Why So Slow?
80. How slow?
81. Benchmark Benchmark.ips do |bm| bm.report 'INDEX: Integration Test' do Minitest.run_one_method(IntegrationTest, 'test_index') end bm.report 'INDEX: Functional Test' do Minitest.run_one_method(ControllerTest, 'test_index') end bm.compare! end Controller Integration Compare
82. Result Calculating ------------------------------------- INDEX: Integration Test 15.000 i/100ms INDEX: Functional Test 39.000 i/100ms ------------------------------------------------- INDEX: Integration Test 158.019 (± 8.2%) i/s - 795.000 INDEX: Functional Test 390.853 (± 9.0%) i/s - 1.950k Comparison: INDEX: Functional Test: 390.9 i/s INDEX: Integration Test: 158.0 i/s - 2.47x slower Comparison
83. What is slow?
84. stackprof
85. Controller Bottleneck Minitest.run_one_method(ControllerTest, 'test_index') StackProf.run(mode: :cpu, out: 'stackprof.dump') do 3000.times do Minitest.run_one_method(ControllerTest, 'test_index') end end
86. CPU time vs Wall time
87. `sleep` is slow, but doesn’t use CPU
88. View Stack $ stackprof stackprof.dump
89. Stack ================================== Mode: cpu(1000) Samples: 3280 (0.00% miss rate) GC: 134 (4.09%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 3146 (95.9%) 1740 (53.0%) Minitest::Runnable.on_signal 483 (14.7%) 483 (14.7%) Dependencies#search_for_file 207 (6.3%) 191 (5.8%) SQLite3::Database#prepare 95 (2.9%) 95 (2.9%) Inflector#underscore 47 (1.4%) 46 (1.4%) TestRequest#initialize
90. 53.0% in Minitest::Runnable.on_signal
91. on_signal def self.on_signal name, action # :nodoc: supported = SIGNALS[name] old_trap = trap name do old_trap.call if old_trap.respond_to? :call action.call end if supported yield ensure trap name, old_trap if supported end
92. JUST DELETE IT
93. on_signal def self.on_signal name, action yield end
94. Results ================================== Mode: cpu(1000) Samples: 4246 (0.00% miss rate) GC: 386 (9.09%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 220 (5.2%) 185 (4.4%) SQLite3::Database#prepare 148 (3.5%) 131 (3.1%) TestRequest#initialize 113 (2.7%) 113 (2.7%) Event#initialize 105 (2.5%) 105 (2.5%) PerThreadRegistry#instance
95. 50% faster!
96. ☺
97. Time the process $ time ruby benchmark.rb real 0m12.358s user 0m11.732s sys 0m0.498s $ time ruby benchmark.rb real 0m12.055s user 0m11.462s sys 0m0.475s Before After
98. &
99. CPU bound
100. Wall should be similar to CPU
101. Stack (CPU) ================================== Mode: cpu(1000) Samples: 3280 (0.00% miss rate) GC: 134 (4.09%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 3146 (95.9%) 1740 (53.0%) Minitest::Runnable.on_signal 483 (14.7%) 483 (14.7%) Dependencies#search_for_file 207 (6.3%) 191 (5.8%) SQLite3::Database#prepare 95 (2.9%) 95 (2.9%) Inflector#underscore 47 (1.4%) 46 (1.4%) TestRequest#initialize
102. Stack (WALL) ================================== Mode: wall(1000) Samples: 8797 (0.11% miss rate) GC: 535 (6.08%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 286 (3.3%) 286 (3.3%) PerThreadRegistry#instance 233 (2.6%) 233 (2.6%) Formatter#non_recursive 336 (3.8%) 227 (2.6%) Class#class_attribute 209 (2.4%) 209 (2.4%) NonConcurrentCacheBackend#[] 198 (2.3%) 198 (2.3%) ActiveRecord::Base.logger
103. '
104. Even profilers have bugs!
105. Something useful
106. Integration test profiles.
107. Stack (WALL) ================================== Mode: wall(1000) Samples: 20168 (2.95% miss rate) GC: 1247 (6.18%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 6401 (31.7%) 5622 (27.9%) block in Module#delegate 495 (2.5%) 495 (2.5%) PerThreadRegistry#instance 341 (1.7%) 341 (1.7%) NonConcurrentCacheBackend#[] 317 (1.6%) 317 (1.6%) Set#include?
108. Calling `delegate` is slow.
109. Patch - delegate :url_for, :optimize_routes_generation?, to: '@_routes' + def url_for(options) + @_routes.url_for(options) + end + + def optimize_routes_generation? + @_routes.optimize_routes_generation? + end +
110. Check our progress
111. 2.47x slower
112. Result Calculating ------------------------------------- INDEX: Integration Test 23.000 i/100ms INDEX: Functional Test 37.000 i/100ms ------------------------------------------------- INDEX: Integration Test 251.487 (±12.7%) i/s - 1.242k INDEX: Functional Test 365.112 (±11.0%) i/s - 1.813k Comparison: INDEX: Functional Test: 365.1 i/s INDEX: Integration Test: 251.5 i/s - 1.45x slower Comparison
113. 2.47x to 1.45x
114. Verify $ time ruby test.rb real 0m26.821s user 0m25.277s sys 0m0.903s $ time ruby test.rb real 0m16.632s user 0m15.709s sys 0m0.638s Before After
115. Even profilers have bugs
116. Always measure
117. Verify results
118. Today: 10% slower
119. But faster than controller tests in previous versions
120. Soup two Nuts Performance
121. bundler
122. Binstubs
123. bin stub source code #!/Users/aaron/.rbenv/versions/ruby-trunk/bin/ruby # # This file was generated by RubyGems. # # The application 'bundler' is installed as part of a gem, and # this file is here to facilitate running it. # # some unimportant stuff happens here! gem 'bundler', version load Gem.bin_path('bundler', 'bundle', version)
124. gem
125. require
126. What does `gem` do?
127. Finds a gemspec
128. Puts a path on $LOAD_PATH
129. Load Path Manipulation > x = $LOAD_PATH.dup; nil => nil > gem 'bundler' => true > $LOAD_PATH - x => ["/Users/aaron/.rbenv/versions/ruby-trunk/lib/ruby/gems/2.3.0/gems/ bundler-1.9.2/lib"]
130. What Happens When File is Required?
131. Is File in $LOAD_PATH? Load it! Is file in Gemspec? Add gemspec to $LOAD_PATH
132. RubyGems performance
133. Test Methodology
134. Require One File And Increase Gems
135. Activate one Gem And Increase Gems
136. time to require one file Time in seconds 0 0.08 0.16 0.24 0.32 Number of Gems on the system 0 100 200 300 400 500 600 700 800 900 1000 best worst
137. allocations to require one file Number of allocations 0 22500 45000 67500 90000 Number of Gems on the system 0 100 200 300 400 500 600 700 800 900 1000 best worst
138. time to activate one gem Time in seconds 0 0.06 0.12 0.18 0.24 Number of Gems on the system 0 100 200 300 400 500 600 700 800 900 1000 best worst
139. allocations to activate one gem Number of allocations 0 10000 20000 30000 40000 Number of Gems on the system 0 100 200 300 400 500 600 700 800 900 1000
140. Bundler Performance
141. Test Method
142. One Gem in Gemfile Increase Gems on System
143. Increase Gems in Gemfile Constant Gems on System
144. *NO DEPENDENCIES
145. *ALL LOCAL
146. Vary System Gems
147. `bundle update` with one gem Time in Seconds 0 0.2 0.4 0.6 0.8 Number of Gems on the system 100 200 300 400 500 600 700 800 900 1000
148. `bundle exec` with one gem Time in Seconds 0 0.2 0.4 0.6 0.8 Number of Gems on the system 100 200 300 400 500 600 700 800 900 1000
149. Vary Gems in Gemfile
150. `bundle exec` time Time in seconds 1.44 1.45 1.46 1.47 1.48 Number of gems in the gemfile 0 100 200 300 400 500 600 700 800 900 1000 y = 0.0008x + 1.4603 R² = 0.1323
151. Bundle Exec Time Breakdown (ms) Other? 750 Spec Load 710
152. `bundle update` time Time in Seconds 0 17.5 35 52.5 70 Number of Gems in the Gemfile 0 100 200 300 400 500 600 700 800 900 1000 y = 0.6984x2 - 2.9466x + 13.856 R² = 0.999
153. Thanks Excel
157. R Code - Fit curve > bundler <- read.csv(file="~/bundle_test/ run_bundle_updats.csv",head=TRUE,sep=",") > fit2 <- lm(bundler$time ~ poly(bundler$gems, 2, raw = TRUE)) > xx <- seq(0, 1000, length=11) > plot(bundler$gems, bundler$time) > lines(xx, predict(fit2, data.frame(x = xx)), col="red")
159. R Code - Coefficients > coefficients(fit2) (Intercept) poly(bundler$gems, 2, raw = TRUE)1 poly(bundler$gems, 2, raw = TRUE)2 1.160827e+01 -1.549806e-02 6.984136e-05 > summary(fit2)$r.squared [1] 0.9989536 >
161. Make Predictions > second_order <- function(newdist, model) { + coefs <- coef(model) + res <- coefs[1] + (coefs[2] * newdist) + (coefs[3] * newdist^2) + return(res) + } > second_order(0, fit2) (Intercept) 11.60827 > second_order(10000, fit2) (Intercept) 6840.763 empty gemfile 10k gemfile
162. How to Fix?
163. What do we know?
164. bin stub source code #!/Users/aaron/.rbenv/versions/ruby-trunk/bin/ruby # # This file was generated by RubyGems. # # The application 'bundler' is installed as part of a gem, and # this file is here to facilitate running it. # # some unimportant stuff happens here! gem 'bundler', version load Gem.bin_path('bundler', 'bundle', version) gem name
165. Gemfile.lock GEM specs: pkg0 (1) PLATFORMS ruby DEPENDENCIES pkg0 gem name and version!
166. Spec loading code today def get_spec name spec_cache = Dir[File.join(dir, "*.gemspec")] spec_cache.find { |spec| spec.name == name } end
167. bundle -v against 100k gems [aaron@TC rubygems (master)]$ time GEM_HOME=`pwd`/100k_gems GEM_PATH=`pwd`/100k_gems ruby -I lib 100k_gems/bin/bundle -v Bundler version 1.9.4 real 0m50.045s user 0m15.815s sys 0m11.967s
168. Spec loading code I want def get_spec name spec_cache = Dir[File.join(dir, "#{name}-*.gemspec")] spec_cache.find { |spec| spec.name == name } end
169. bundle -v against 100k gems [aaron@TC rubygems (specs)]$ time GEM_HOME=`pwd`/100k_gems GEM_PATH=`pwd`/100k_gems ruby -I lib 100k_gems/bin/bundle -v Bundler version 1.9.4 real 0m0.632s user 0m0.479s sys 0m0.124s
170. Still O(n)
171. I think we can make it O(1)
172. Future Work: Make `require` O(1)
173. Don’t recalculate static data
174. Take measurements
175. Question intuitions
176. I have stickers!
177. THANK YOU!!!