Improving CVAR Performance in Ruby 3.1

Have you ever wondered how class variables (CVARs) in Ruby work? Would you be surprised to learn that their performance becomes increasingly worse as the inheritance chain grows? I’m excited to share that in Ruby 3.1 we fixed the performance of CVARs.

In this talk we'll look at the language design of class variables, learn about how they work, and, deep dive into how we improved their performance in Ruby 3.1 by adding a cache! We'll look at the cache design and real-world benchmarks showing how much faster they are regardless of the size of the inheritance chain.

Transcript

1. Improving CVAR performance in Ruby 3.1 EILEEN M. UCHITELLE | @eileencodes
2. Thank You RubyConf organizers, staff, & attendees —
3. Eileen M. Uchitelle @eileencodes
4. Core Team Member of Ruby on Rails
5. f Principal Software Engineer at GitHub
6. Improving CVAR performance in Ruby 3.1
7. What is a CVAR?
8. What is a CVAR? class MyClass @@cvar = 1 def self.read_cvar @@cvar end end
9. What is a CVAR? class MyClass @@cvar = 1 def self.read_cvar @@cvar end end Writer
10. What is a CVAR? class MyClass @@cvar = 1 def self.read_cvar @@cvar end end Reader
11. How does CVAR inheritance work?
12. IVAR inheritance class Dog attr_reader :name, :owner def initialize @name = "Arya" @owner = "Eileen" end end class Puppy < Dog def initialize @name = "Sansa" end end
13. IVAR inheritance > Dog.new.name => "Arya" > Dog.new.owner => "Eileen" > Puppy.new.name => "Sansa" > Puppy.new.owner => nil
14. CVAR inheritance class Dog @@name = "Arya" @@owner = "Eileen" def self.name @@name end def self.owner @@owner end end class Puppy < Dog @@name = "Sansa" end
15. CVAR inheritance > Dog.owner => "Eileen" > Puppy.owner => "Eileen"
16. CVAR inheritance > Dog.owner => "Eileen" > Puppy.owner => "Eileen" > Dog.name => "Sansa"
17. CVAR inheritance > Dog.owner => "Eileen" > Puppy.owner => "Eileen" > Dog.name => "Sansa" > Puppy.name => "Sansa" 🤔
18. Puppy Object Dog Kernel BasicObject
19. Puppy Dog Object Kernel BasicObject Stores @@name @@owner
20. @@name "Sansa" RCLASS_IV_TBL Puppy Dog
21. Understanding CVAR Overtaken
22. CVAR Overtaken class Dog end class Puppy < Dog @@name = "Sansa" def self.name @@name end end
23. CVAR Overtaken class Dog end class Puppy < Dog @@name = "Sansa" def self.name @@name end end class Dog @@name = "Arya" end
24. CVAR Overtaken > Puppy.name => class variable @@name of Puppy is overtaken by Dog
25. The deeper the inheritance chain the slower CVAR access is
26. CVAR Benchmarks MODULES = ["A", ..."WWWW"] class One @@cvar = 1 def self.cvar @@cvar end eval <<-EOM module #{MODULES.first} end include #{MODULES.first} EOM end
27. CVAR Benchmarks class Thirty @@cvar = 1 def self.cvar @@cvar end MODULES.take(30).each do |module_name| eval <<-EOM module #{module_name} end include #{module_name} EOM end end
28. CVAR Benchmarks class OneHundred @@cvar = 1 def self.cvar @@cvar end MODULES.each do |module_name| eval <<-EOM module #{module_name} end include #{module_name} EOM end end
29. CVAR Benchmarks Benchmark.ips do |x| x.report "1 module" do One.cvar end x.report "30 modules" do Thirty.cvar end x.report "100 modules" do OneHundred.cvar end x.compare! end
30. CVAR Benchmarks Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower
31. S CVAR Benchmarks Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower 8.5x!
32. Are CVARs really that common?
33. CVARs in Rails module ActiveRecord module Core def self.configurations=(config) @@configurations = ActiveRecord::DatabaseConfigs.new(config) end self.configurations = {} def self.configurations @@configurations end end end
34. CVARs in Rails module ActiveRecord module Core mattr_accessor :logger, instance_writer: false mattr_accessor :verbose_query_logs, instance_writer: false, default: false mattr_accessor :schema_format, instance_writer: false, default: :ruby mattr_accessor :dump_schemas, instance_writer: false, default: :schema_search_path mattr_accessor :dump_schema_after_migration, instance_writer: false, default: true end end
35. CVARs in Rails > ActiveRecord::Base.ancestors.count => 73 > ActionController::Base.ancestors.count => 71 > ActiveJob.ancestors.count => 29
36. Building a cache for CVARs
38. @@name "Sansa" RCLASS_IV_TBL RCLASS_CVC_TBL Dog
39. @@name "Sansa" RCLASS_IV_TBL @@name RCLASS_CVC_TBL Dog
40. @@name "Sansa" RCLASS_IV_TBL @@name Inline cache RCLASS_CVC_TBL Dog
41. @@name "Sansa" RCLASS_IV_TBL @@name Inline cache RCLASS_CVC_TBL Dog pointer global_cvar_state Dog
42. @@name "Sansa" RCLASS_IV_TBL @@name Inline cache RCLASS_CVC_TBL Dog pointer global_cvar_state Dog Puppy
43. Puppy.name
44. Puppy.name global_cvar_state == GET_GLOBAL_CVAR_STATE()
45. Puppy.name Dog global_cvar_state == GET_GLOBAL_CVAR_STATE()
46. Dog Puppy.name global_cvar_state == GET_GLOBAL_CVAR_STATE() @@name "Sansa" RCLASS_IV_TBL
47. S CVAR Cache PR github.com/ruby/ruby/pull/4340
48. Benchmarking performance
49. S CVAR Benchmarks Warming up -------------------------------------- 1 module 1.231M i/100ms 30 modules 432.020k i/100ms 100 modules 145.399k i/100ms Calculating ------------------------------------- 1 module 12.210M (± 2.1%) i/s - 61.553M in 5.043400s 30 modules 4.354M (± 2.7%) i/s - 22.033M in 5.063839s 100 modules 1.434M (± 2.9%) i/s - 7.270M in 5.072531s Comparison: 1 module: 12209958.3 i/s 30 modules: 4354217.8 i/s - 2.80x (± 0.00) slower 100 modules: 1434447.3 i/s - 8.51x (± 0.00) slower 8.5x!
50. S CVAR Benchmarks Warming up -------------------------------------- 1 module 1.641M i/100ms 30 modules 1.655M i/100ms 100 modules 1.620M i/100ms Calculating ------------------------------------- 1 module 16.279M (± 3.8%) i/s - 82.038M in 5.046923s 30 modules 15.891M (± 3.9%) i/s - 79.459M in 5.007958s 100 modules 16.087M (± 3.6%) i/s - 81.005M in 5.041931s Comparison: 1 module: 16279458.0 i/s 100 modules: 16087484.6 i/s 30 modules: 15891406.2 i/s No difference!
51. S CVAR Benchmarks $ RAILS_ENV=production INTERVAL=100 WARMUP=1 BENCHMARK=10000 ruby bin/bench ruby 3.1.0dev (2021-06-04T00:24:57Z master 91c542ad05) [x86_64-darwin19] Warming up... Warmup: 1 requests Benchmark: 10000 requests Request per second: 615.1 [#/s] (mean) Percentage of the requests served within a certain time (ms) 50% 1.57 66% 1.68 75% 1.74 80% 1.78 90% 1.91 95% 2.06 98% 2.36 99% 2.67 100% 35.15
52. S CVAR Benchmarks $ RAILS_ENV=production INTERVAL=100 WARMUP=1 BENCHMARK=10000 ruby bin/bench ruby 3.1.0dev (2021-06-04T17:40:20Z add-cache-for.. 37c96af98b) [x86_64-darwin19] Warming up... Warmup: 1 requests Benchmark: 10000 requests Request per second: 657.1 [#/s] (mean) Percentage of the requests served within a certain time (ms) 50% 1.46 66% 1.56 75% 1.63 80% 1.68 90% 1.82 95% 2.01 98% 2.28 99% 2.50 100% 35.13
53. What did we learn?
54. 🐞!
55. S Feature Request
56. Every OSS change has tradeoffs
57. Every OSS change has tradeoffs
58. Tradeoffs Increased complexity
59. Tradeoffs Encourages more usage
60. Tradeoffs Increased maintenance burden
61. Every OSS change is a negotiation
62. Negotiation CVARs aren't going anywhere
63. Negotiation Demonstrate real-world improvements
64. Negotiation Improve all applications by upgrading
65. Let's do more of this
66. Let's do more of this By learning C
67. Let's do more of this By reading "Ruby Under a Microscope"
68. Let's do more of this By making small changes
69. Making Ruby better benefits everyone
70. Making Ruby better benefits you
71. Let's go make Ruby better, together
72. Thank You RubyConf! EILEEN M. UCHITELLE | @eileencodes