DEV: add debugging scripts for memory leaks

These scripts are somewhat rough but I needed them to help debug a memory
leak we have noticed in rails 6.

The biggest object script finds all the biggest objects we have in memory
after boot.

The test memory leak runs a very simple iteration through all multisites
and observed memory.
This commit is contained in:
Sam Saffron 2019-10-03 16:36:16 +10:00
parent 45a55a3abf
commit 038a38ae1c
2 changed files with 144 additions and 0 deletions

36
script/biggest_objects.rb Normal file
View File

@ -0,0 +1,36 @@
# frozen_string_literal: true
# simple script to measure largest objects in memory post boot
if ENV['RAILS_ENV'] != "production"
exec "RAILS_ENV=production ruby #{__FILE__}"
end
require 'objspace'
ObjectSpace.trace_object_allocations do
require File.expand_path("../../config/environment", __FILE__)
Rails.application.routes.recognize_path('abc') rescue nil
# load up the yaml for the localization bits, in master process
I18n.t(:posts)
RailsMultisite::ConnectionManagement.each_connection do
(ActiveRecord::Base.connection.tables - %w[schema_migrations versions]).each do |table|
table.classify.constantize.first rescue nil
end
end
end
5.times do
GC.start(full_mark: true, immediate_sweep: true)
end
[String, Array, Hash].each do |klass|
ObjectSpace.each_object(klass).sort { |a, b| b.length <=> a.length }.first(50).each do |obj|
puts "#{klass} size: #{obj.length} #{ObjectSpace.allocation_sourcefile(obj)} #{ObjectSpace.allocation_sourceline(obj)}"
end
end

108
script/test_memory_leak.rb Normal file
View File

@ -0,0 +1,108 @@
# frozen_string_literal: true
# simple test to check for memory leaks
#
# this performs a trivial operation walking all multisites and grabbing first topic / localizing
# the expectation is that RSS will remain static no matter how many iterations run
if ENV['RAILS_ENV'] != "production"
exec "RAILS_ENV=production ruby #{__FILE__}"
end
if !ENV['LD_PRELOAD']
exec "LD_PRELOAD=/usr/lib/libjemalloc.so.1 ruby #{__FILE__}"
end
if ENV['LD_PRELOAD'].include?("jemalloc")
# for 3.6.0 we need a patch jemal 1.1.0 gem (1.1.1 does not support 3.6.0)
# however ffi is a problem so we need to patch the gem
require 'jemal'
$jemalloc = true
end
if ENV['LD_PRELOAD'].include?("mwrap")
$mwrap = true
require 'mwrap'
end
def bin_diff(current)
$baseline[:arenas].each_with_index do |arena, i|
next if !arena || !arena[:bins]
arena[:bins].each do |size, stat|
allocated = (current.dig(:arenas, i, :bins, size, :allocated) || 0)
diff = allocated - stat[:allocated]
puts "bin #{size} delta #{diff}"
end
end
end
require File.expand_path("../../config/environment", __FILE__)
Rails.application.routes.recognize_path('abc') rescue nil
I18n.t(:posts)
def rss
`ps -o rss -p #{$$}`.chomp.split("\n").last.to_i
end
def loop_sites
RailsMultisite::ConnectionManagement.each_connection do
yield
end
end
def biggest_klass(klass)
ObjectSpace.each_object(klass).max { |a, b| a.length <=> b.length }
end
def iter(warmup: false)
loop_sites { Topic.first; I18n.t('too_late_to_edit') }
if !warmup
GC.start(full_mark: true, immediate_sweep: true)
if $jemalloc
jemal_stats = Jemal.stats
jedelta = "(jdelta #{jemal_stats[:active] - $baseline_jemalloc_active})"
end
if $mwrap
mwrap_delta = (Mwrap.total_bytes_allocated - Mwrap.total_bytes_freed) - $mwrap_baseline
mwrap_delta = "(mwrap delta #{mwrap_delta})"
end
rss_delta = rss - $baseline_rss
array_delta = biggest_klass(Array).length - $biggest_array_length
puts "rss: #{rss} (#{rss_delta}) #{mwrap_delta}#{jedelta} heap_delta: #{GC.stat[:heap_live_slots] - $baseline_slots} array_delta: #{array_delta}"
if $jemalloc
bin_diff(jemal_stats)
end
end
end
iter(warmup: true)
4.times do
GC.start(full_mark: true, immediate_sweep: true)
end
if $jemalloc
$baseline = Jemal.stats
$baseline_jemalloc_active = $baseline[:active]
4.times do
GC.start(full_mark: true, immediate_sweep: true)
end
end
if $mwrap
$mwrap_baseline = Mwrap.total_bytes_allocated - Mwrap.total_bytes_freed
end
$baseline_slots = GC.stat[:heap_live_slots]
$baseline_rss = rss
$biggest_array_length = biggest_klass(Array).length
100000.times do
iter
end