Refactor ActiveSupport subscriber

[kaylareopelle]

This PR is an experiment to try to speed up the ActiveSupport instrumentation.

Where strings were used, constants are now used instead.

There's a small improvement to use match? + byteslice over =~ + $1

match? performance benchmark:

# frozen_string_literal: true

require 'bundler/inline'

gemfile do
  source 'https://rubygems.org'
  gem 'benchmark-ips'
  gem 'benchmark-ipsa'
end

PATTERN = /\Acache_([^\.]*)\.active_support\z/

original_query_names = %w[
  cache_read.active_support
  cache_read_multi.active_support
  cache_generate.active_support
  cache_fetch_hit.active_support
  cache_write.active_support
  cache_write_multi.active_support
  cache_increment.active_support
  cache_decrement.active_support
  cache_delete.active_support
  cache_delete_multi.active_support
  cache_delete_matched.active_support
  cache_cleanup.active_support
  cache_prune.active_support
  cache_exist?.active_support
]

def equal_tilde(k)
  PATTERN =~ k
end

def matches?(k)
  PATTERN.match?(k)
end

def matchy_match(k)
  PATTERN.match(k) do |m|
    true
  end
end

OOPS = 'oops'

Benchmark.ipsa do |x|
  x.report('=~')         { original_query_names.each { |n| raise OOPS unless equal_tilde(n) } }
  x.report('match?') { original_query_names.each { |n| raise OOPS unless matches?(n) } }
  x.report('match')   { original_query_names.each { |n| raise OOPS unless matchy_match(n) } }
  x.compare!
end


# Allocations -------------------------------------
#                   =~      15/0  alloc/ret        0/0  strings/ret
#               match?       0/0  alloc/ret        0/0  strings/ret
#                match      14/0  alloc/ret        0/0  strings/ret
# Warming up --------------------------------------
#                   =~    10.294k i/100ms
#               match?    26.665k i/100ms
#                match    10.279k i/100ms
# Calculating -------------------------------------
#                   =~    119.819k (± 9.7%) i/s -    597.052k
#               match?    282.765k (± 9.0%) i/s -      1.413M
#                match    105.432k (± 8.1%) i/s -    524.229k

# Comparison:
#               match?:   282765.3 i/s
#                   =~:   119818.9 i/s - 2.36x slower
#                match:   105432.3 i/s - 2.68x slower

Comparing match? with byteslice alongside =~ with $1

# frozen_string_literal: true

require 'bundler/inline'

gemfile do
  source 'https://rubygems.org'
  gem 'benchmark-ips'
  gem 'benchmark-ipsa'
end

PATTERN = /\Acache_([^\.]*)\.active_support\z/

NAMES = %w[
  cache_read.active_support
  cache_read_multi.active_support
  cache_generate.active_support
  cache_fetch_hit.active_support
  cache_write.active_support
  cache_write_multi.active_support
  cache_increment.active_support
  cache_decrement.active_support
  cache_delete.active_support
  cache_delete_multi.active_support
  cache_delete_matched.active_support
  cache_cleanup.active_support
  cache_prune.active_support
  cache_exist?.active_support
]

Benchmark.ipsa do |x|
  x.report("match? with byteslice") { NAMES.each { |k| PATTERN.match?(k) ? k.byteslice(6..-16) : UNKNOWN } }
  x.report("=~ with $1")            { NAMES.each { |k| PATTERN =~ k ? $1 : UNKNOWN } }

  x.compare!
end

# Allocations -------------------------------------
# match? with byteslice
#                           15/0  alloc/ret       14/0  strings/ret
#           =~ with $1      28/1  alloc/ret       14/0  strings/ret
# Warming up --------------------------------------
# match? with byteslice
#                         18.999k i/100ms
#           =~ with $1    10.607k i/100ms
# Calculating -------------------------------------
# match? with byteslice
#                         203.823k (±10.6%) i/s -      1.007M
#           =~ with $1    101.897k (±18.1%) i/s -    498.529k

# Comparison:
# match? with byteslice:   203823.3 i/s
#           =~ with $1:   101896.7 i/s - 2.00x slower

[github-actions]

SimpleCov Report

	Coverage	Threshold
Line	93.77%	93%
Branch	71.48%	71%

[kaylareopelle]

We're running a different test to improve the subscriber. See #2610.