Skip to content

rbUUbr/understand_ruby_memory

BranchesTags
 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

41 Commits
1.txt
1.txt
 
 
2.txt
2.txt
 
 
3.txt
3.txt
 
 
README.md
README.md
 
 
memory_freeze_benchmark.rb
memory_freeze_benchmark.rb
 
 

Repository files navigation

Understand ruby memory usage 🤔

I initially made this post to add questions I have about ruby memory (in MRI). I don't have any CS degree, never did some C, understanding how ruby use memory is not an easy path. I'm passionate about this subject.

I will add more questions in the future using pull requests so feel free to watch the repo.

Make a PR if you want to add links, answer to a question, or simply correct what I said. I would love that. ✨

Questions

Does an object that is allocated can turn to be retained because he is still present after few GC [Answered] ?

From ("retained" vs "allocated" from memory_profiler README)

Yes. The term "allocated" vs "retained" may vary depending on the tool you use, but, for example, in memory_profiler, "retained" means "survived a major GC".

What are the first line of a heap dump that are not address [Answered] ?

Header of the heap dump from heapy gem (carefull is 78mo text file)

{"type":"ROOT", "root":"vm", "references":["0x7fb4740bc400", "0x7fb4740b79a0", "0x7fb4740dff68", "0x7fb4740bff60", "0x7fb4740bff10", "0x7fb474c13a88", "0x7fb474ac0618", "0x7fb4740bfe98", "0x7fb4740bfe70", "0x7fb4740ddc68", "0x7fb4740dff90", "0x7fb4772f88d8", "0x7fb4772f8900"]}
{"type":"ROOT", "root":"finalizers", "references":["0x7fb4768584c8", "0x7fb474f18a58", "0x7fb477083ad0", "0x7fb4739cd040", "0x7fb4772c99c0", "0x7fb475fa9188", "0x7fb475e00368", "0x7fb475d99320"]}
{"type":"ROOT", "root":"machine_context", "references":["0x7fb474027760", "0x7fb474027350", "0x7fb4740bed18", "0x7fb4740becc8", "0x7fb4740bed18", "0x7fb4772f8900", "0x7fb4740ddc68", "0x7fb474027850", "0x7fb474027850", "0x7fb4740becc8", "0x7fb4740becc8", "0x7fb4740becc8", "0x7fb475dc0ab0", "0x7fb476ab2458", "0x7fb4740bfd58", "0x7fb4740bfd58", "0x7fb4740bfd58", "0x7fb4740ddc68", "0x7fb475dc0b78", "0x7fb4768919f8", "0x7fb4740ddc68", "0x7fb4738e33c8", "0x7fb4740ddc68", "0x7fb4740ddc68", "0x7fb4740ddc68", "0x7fb4740dec58", "0x7fb4740dec58", "0x7fb4740ddc68", "0x7fb4740dec58", "0x7fb475dc26d0", "0x7fb475dc26d0", "0x7fb475dc26d0", "0x7fb4740bfd58", "0x7fb475dc26d0", "0x7fb475dc0808", "0x7fb47587d828", "0x7fb4740deca8", "0x7fb4740bfd58", "0x7fb4740bfd58", "0x7fb4740bfd58", "0x7fb4738e33c8", "0x7fb4738ed0d0", "0x7fb475dc23b0", "0x7fb475dc0b78", "0x7fb475dc23b0", "0x7fb475dc23b0", "0x7fb4772f99b8", "0x7fb475dc23b0", "0x7fb4772f99b8", "0x7fb475dc2360", "0x7fb476862d38", "0x7fb475dc26d0", "0x7fb475dc2360", "0x7fb4740cd980", "0x7fb475dc2360", "0x7fb4772f9800", "0x7fb4772f9828", "0x7fb4772f9828", "0x7fb4740dced0", "0x7fb4740cd980", "0x7fb4740cd980", "0x7fb4740cd980", "0x7fb475dc23b0", "0x7fb476862d38", "0x7fb475dc0f88", "0x7fb475de8268", "0x7fb475de83f8", "0x7fb4740ddc68", "0x7fb4740dec58", "0x7fb475de8268", "0x7fb475de83d0", "0x7fb475df1bb0", "0x7fb475de83f8", "0x7fb475df1bb0", "0x7fb4740ddc68", "0x7fb475df1c50", "0x7fb4740ddc68", "0x7fb475df1bd8", "0x7fb4740ddc68", "0x7fb4740deaf0", "0x7fb4740ddc68", "0x7fb476988a28", "0x7fb476988a00", "0x7fb476988d98", "0x7fb476988d98", "0x7fb476988a00", "0x7fb4740deaf0", "0x7fb4740ddc40", "0x7fb4740ddc40", "0x7fb4740bc338", "0x7fb47490a2b0", "0x7fb4740ddc68", "0x7fb47698ac88", "0x7fb47698ac88", "0x7fb47698ac88", "0x7fb47698ac88", "0x7fb47698ac88"]}
{"type":"ROOT", "root":"global_list", "references":["0x7fb4759d3678", "0x7fb4759d3768", "0x7fb4759d3790", "0x7fb4759d37e0", "0x7fb4759d3808", "0x7fb4759d3830", "0x7fb4759d38d0", "0x7fb4759d3920", "0x7fb4759d3970", "0x7fb4759d3998", "0x7fb4759d39e8", "0x7fb4759d3a10", "0x7fb4759d3a38", "0x7fb4759d3a88", "0x7fb4759d3ab0", "0x7fb4759d3ad8", "0x7fb4759d3b00", "0x7fb4759d3b28", "0x7fb4759d3b78", "0x7fb4759d3ba0", "0x7fb4759d3bc8", "0x7fb4759d3bf0", "0x7fb4759d3c18", "0x7fb4759d3c40", "0x7fb4759d3c68", "0x7fb4759d3c90", "0x7fb4759d3cb8", "0x7fb4759d3ce0", "0x7fb4759d3d58", "0x7fb4759d3dd0", "0x7fb4759d3df8", "0x7fb4759d3e20", "0x7fb4759d3e70", "0x7fb4759d3ee8", "0x7fb4759d3f10", "0x7fb4759d3fb0", "0x7fb474135fa8", "0x7fb4741353c8", "0x7fb474134270", "0x7fb474134298", "0x7fb4741342c0", "0x7fb4741342e8", "0x7fb474134310", "0x7fb474134338", "0x7fb474134360", "0x7fb474134388", "0x7fb4741343b0", "0x7fb4741343d8", "0x7fb474134428", "0x7fb474134450", "0x7fb474134478", "0x7fb4741344a0", "0x7fb4741344c8", "0x7fb4741344f0", "0x7fb474134518", "0x7fb474134540", "0x7fb474134568", "0x7fb474134590", "0x7fb474134608", "0x7fb47481da50", "0x7fb47481dac8", "0x7fb47481db40", "0x7fb47481db68", "0x7fb47481dd20", "0x7fb47481dd98", "0x7fb47481dfa0", "0x7fb47481dfc8", "0x7fb47481e018", "0x7fb47481e040", "0x7fb47481e248", "0x7fb47481e388", "0x7fb47481e5e0", "0x7fb47481e608", "0x7fb47481e6a8", "0x7fb47481e770", "0x7fb47481e7e8", "0x7fb47481e8d8", "0x7fb47481e928", "0x7fb47481e9c8", "0x7fb47481ea40", "0x7fb47481eae0", "0x7fb47481ec48", "0x7fb47481ed88", "0x7fb47481ee28", "0x7fb47481ee50", "0x7fb47481f058", "0x7fb47481f0d0", "0x7fb47481f0f8", "0x7fb47481f210", "0x7fb47481f288", "0x7fb47481f378", "0x7fb47481f418", "0x7fb47481f440", "0x7fb47481f4b8", "0x7fb47481f508", "0x7fb47481f710", "0x7fb47481f738", "0x7fb47481f760", "0x7fb47481fad0", "0x7fb47481fb20", "0x7fb47481fb70", "0x7fb47481fbe8", "0x7fb47584f270", "0x7fb475874700", "0x7fb475874a20", "0x7fb4769904d0", "0x7fb4740dff40"]}

Answer by Aaron Patterson benoittgt#1

Those are "ROOTS" (the ones with type = "ROOTS"). The system has various "roots".

What is a "ROOT"?

Objects in the system form a tree data structure. Imagine some code like this:

a = "foo"
b = "bar"
c = { a => b }

It's easy to see how c is connected to a and b and prevents them from being garbage collected:

abc

But say we have this:

a = "foo"
b = "bar"
c = { a => b }
GC.start
p c

What prevents c from being garbage collected? Something must hold a reference to c so that the garbage collector can know that it shouldn't be GC'd. This is where ROOTS come in. Roots are special nodes in the system that hold on to these references:

roots

MRI has a few different places it considers a root, and those are in the name root (like vm, finalizers, etc). Explaining each of these types is a little too long for here, but you can think of them as the root of the tree that forms your object graph. They keep your program objects alive, and they are the starting point for the GC to find live objects in your system. In fact the graph I showed above is a little inaccurate. Since a and b are also top level local variables like c, the graph looks a bit more like this:

total

Hope that helps!

For the rest of the dump. Go to https://blog.codeship.com/the-definitive-guide-to-ruby-heap-dumps-part-i/ but not the header.

Manually inspecting this file might be of some interest, but we really need to aggregate information to make use of this data. Before we do that, let’s look at some of the keys in the generated JSON.

  • generation: The garbage collection generation where the object was generated
  • file: The file where the object was generated
  • line: The line number where the object was generated
  • address: This is the memory address of the object
  • memsize: The amount of memory the object consumes
  • references: The memory addresses of other objects that this object retains There are other keys, but that’s enough for now. It’s worth noting that several of these are optional. For example if an object was generated before you started tracing object allocations, it won’t contain generation, file, or line information.

What is allocated and what is not allocated (Not every object requires allocation) [Answered] ?

From Aaron Patterson - Defragging Ruby, RubyConfBY 2017 talk. Aaron Patterson point to gc.c file from ruby source code, and look at object_id.

 *  Note: that some objects of builtin classes are reused for optimization.
 *  This is the case for immediate values and frozen string literals.
 *
 *  Immediate values are not passed by reference but are passed by value:
 *  +nil+, +true+, +false+, Fixnums, Symbols, and some Floats.
 *
 *      Object.new.object_id  == Object.new.object_id  # => false
 *      (21 * 2).object_id    == (21 * 2).object_id    # => true
 *      "hello".object_id     == "hello".object_id     # => false
 *      "hi".freeze.object_id == "hi".freeze.object_id # => true

And from an email of Koichi Sasada.

Ruby has internal types and all objects are belong to only one type.

The followings are all types. from include/ruby/ruby.h

#define T_NONE   RUBY_T_NONE
#define T_NIL    RUBY_T_NIL
#define T_OBJECT RUBY_T_OBJECT
#define T_CLASS  RUBY_T_CLASS
#define T_ICLASS RUBY_T_ICLASS
#define T_MODULE RUBY_T_MODULE
#define T_FLOAT  RUBY_T_FLOAT
#define T_STRING RUBY_T_STRING
#define T_REGEXP RUBY_T_REGEXP
#define T_ARRAY  RUBY_T_ARRAY
#define T_HASH   RUBY_T_HASH
#define T_STRUCT RUBY_T_STRUCT
#define T_BIGNUM RUBY_T_BIGNUM
#define T_FILE   RUBY_T_FILE
#define T_FIXNUM RUBY_T_FIXNUM
#define T_TRUE   RUBY_T_TRUE
#define T_FALSE  RUBY_T_FALSE
#define T_DATA   RUBY_T_DATA
#define T_MATCH  RUBY_T_MATCH
#define T_SYMBOL RUBY_T_SYMBOL
#define T_RATIONAL RUBY_T_RATIONAL
#define T_COMPLEX RUBY_T_COMPLEX
#define T_IMEMO  RUBY_T_IMEMO
#define T_UNDEF  RUBY_T_UNDEF
#define T_NODE   RUBY_T_NODE
#define T_ZOMBIE RUBY_T_ZOMBIE

We call types as T_xxx.

Internal:
   T_NONE (reserved, not used)
   T_DATA
   T_IMEMO
   T_NODE
   T_ZOMBIE

Not allocated:
   T_NIL (nil)
   T_FIXNUM (small integers)
   T_TRUE (true)
   T_FALSE (false)
   T_UNDEF (internal use)

Allocated (normal)
   T_OBJECT
   T_CLASS
   T_ICLASS (internal use, to implement Module#include)
   T_MODULE
   T_FLOAT (now Ruby has flonum tech so that most of Float values
            are not allocated on 64 bit CPU)
   T_STRING
   T_REGEXP
   T_ARRAY
   T_HASH
   T_STRUCT
   T_BIGNUM (Integer representation)
   T_FILE
   T_MATCH
   T_SYMBOL (there are two symbols: allocated or not allocated)
   T_RATIONAL
   T_COMPLEX

What is garbage collected [Not Answered] ?

What is garbage collected or not ?

🐕 garbage collected?
integers no because not allocated?
booleans no because not allocated?
nil no because not allocated?
strings yes
symbols it depends?
constants If you remove a constant, the object it points to will be GC'd. But constants aren't really a type of object, it's just a name. source
freeze objects yes
arrays yes?
hashes yes?
what else? ?

Why people are always scared about time spent in GC when the Newrelic graph of my app show an average time spent in GC that is 0.0676% ?

Nate Berkopec's opinion - most people assume garbage collection takes a lot of time in a GC'd language. GC languages are slower than non-GC'd languages, therefore it must be GC that is slow. However, it's not just GC but allocation and the record keeping that goes along with it that slows Ruby programs. Creating an object isn't just the simple process of writing to some memory location, we have to do a bunch of checks, look for free space in the ObjectSpace, etc etc. So a GC'd language is slower, but we actually incur most of the cost while running the program, not when garbage collecting.

Why when using a frozen string we don't allocate memory [Answered] ?

I use a method because it represents "patterns" we discuss with my team, I try to measure the number of allocations betweens calling directly string, calling a string set into a constant outside the function and calling a string frozen in a constant outside the function.

require 'memory_profiler'

report_1 = MemoryProfiler.report do
  def get_me_directly
    "hey"
  end
  get_me_directly
end

report_2 = MemoryProfiler.report do
  ST = "yep"
  def get_me_with_constant
    ST
  end
  get_me_with_constant
end

report_3 = MemoryProfiler.report do
  ST_FREEZE = "yop".freeze
  def get_me_with_constant_freeze
    ST_FREEZE
  end
  get_me_with_constant_freeze
end

report_1.pretty_print
# Allocated String Report
# -----------------------------------
#          1  "hey"
#          1  measure_allocation.rb:5
#
#
# Retained String Report
# -----------------------------------

report_2.pretty_print
# Allocated String Report
# -----------------------------------
#          1  "yep"
#          1  measure_allocation.rb:11
#
#
# Retained String Report
# -----------------------------------
#          1  "yep"
#          1  measure_allocation.rb:11

report_3.pretty_print
# Allocated String Report
# -----------------------------------
#
# Retained String Report
# -----------------------------------

As mentionned by Sam Saffron it can happen when :

it was allocated earlier when the ruby code was parsed. You can confirm this by loading a file and putting the load in a mem profiler block.

Check and updated benchmark that doesn't show this weird behavior : https://github.com/benoittgt/understand_ruby_memory/blob/master/memory_freeze_benchmark.rb

Why generation number in heap dump are in random order [Answered] ?

When you read heap dump you have lines like this :

{"address":"0x7fb27108e3e8", "type":"IMEMO", "class":"0x7fb26aa79cd8", "references":["0x7fb26aa7aa20", "0x7fb26aa7b0b0", "0x7fb26aa7a9d0"], "file":"/Users/bti/.rvm/gems/ruby-2.3.0/gems/activerecord-4.2.5.2/lib/active_record/attributes.rb", "line":106, "method":"reset_column_information", "generation":53, "memsize":40, "flags":{"wb_protected":true, "old":true, "uncollectible":true, "marked":true}}

We can read on How I spent two weeks hunting a memory leak that:

you can enable object creation tracing, and for each newly created object the location where it was instantiated (source file, line) will be saved and accessible later.

I generated a heap dump from a ruby program... and extract generation number in the dump with jq.

$ cat tmp/2017-04-10T19:32:45+02:00-heap.dump | jq 'select(.generation != null) | "\(.generation) "' -j
... 57 57 57 57 58 51 58 58 58 58 58 51 58 58 58 58 58 58 51 58 58 58 58 58 51 58 51 51 51 51 51 51 51 51 51 51 ...

As you can see generation number are in random order. After watching Methods of Memory Management in MRI by Aaron Patterson it's a normal behavior to realocate in other generation.

In order to classify objects as new or old the GC does the following: whenever it marks an object in a mark-phase (which means that the object will survive this GC run) it promotes it to the old generation. Watching and Understanding the Ruby 2.1 Garbage Collector at Work

A generational GC (also known as ephemeral GC) divides objects into generations and, on most cycles, will place only the objects of a subset of generations into the initial white (condemned) set. Tracing garbage collection

As Koichi said :

Generation is an age of an object. If you measure ages of people walking in a street, it will be random numbers.

Why keywords arguments will create less garbage collected objects after 2.2 [Answered] ?

I asked Akira Matsuda after watching Ruby 2 in Ruby on Rails - RedDotRubyConf 2017 about something he mentions:

Ruby 2 kwargs ... create less garbage collected objects (maybe)

His answer is a gist with an allocation_tracer measure between ruby 2.1.10 and ruby 2.2.7 on different methods with different "type" of params in method.

require 'allocation_tracer'
require 'active_support/core_ext/array/extract_options'
require 'pp'

# Active Support
def foo1(options = {})
  options[:x] || 1
end

# Ruby
def foo2(x: 1)
  x
end

# Active Support, varargs
def bar1(*args)
  options = args.extract_options!
  options[:y] || 1
end

# Ruby, varargs
def bar2(*args, y: 1)
  y
end

ObjectSpace::AllocationTracer.setup(%i{path line type})


foo1 x: 2
foo2 x: 2
bar1 y: 2
bar2 y: 2

pp activesupport_1: ObjectSpace::AllocationTracer.trace {
  foo1 x: 2
}
pp kwargs_1: ObjectSpace::AllocationTracer.trace {
  foo2 x: 2
}

pp activesupport_2: ObjectSpace::AllocationTracer.trace {
  bar1 y: 2
}
pp kwargs_2: ObjectSpace::AllocationTracer.trace {
  bar2 y: 2
}

That returns

# ruby 2.1.10p492 (2016-04-01 revision 54464) [x86_64-darwin15.0]

{:activesupport_1=>{["kwargs_allocation.rb", 31, :T_HASH]=>[1, 0, 0, 0, 0, 0]}}
{:kwargs_1=>{["kwargs_allocation.rb", 34, :T_HASH]=>[2, 0, 0, 0, 0, 0]}}
{:activesupport_2=>
  {["kwargs_allocation.rb", 38, :T_HASH]=>[1, 0, 0, 0, 0, 0],
   ["kwargs_allocation.rb", 38, :T_ARRAY]=>[1, 0, 0, 0, 0, 0]}}
{:kwargs_2=>
  {["kwargs_allocation.rb", 41, :T_HASH]=>[2, 0, 0, 0, 0, 0],
   ["kwargs_allocation.rb", 41, :T_ARRAY]=>[1, 0, 0, 0, 0, 0]}}

# ruby 2.2.7p470 (2017-03-28 revision 58194) [x86_64-darwin15]..ruby 2.5.0dev (2017-05-19 trunk 58790) [x86_64-darwin15]

{:activesupport_1=>{["kwargs_allocation.rb", 31, :T_HASH]=>[1, 0, 0, 0, 0, 0]}}
{:kwargs_1=>{}}
{:activesupport_2=>
  {["kwargs_allocation.rb", 38, :T_HASH]=>[1, 0, 0, 0, 0, 0],
   ["kwargs_allocation.rb", 38, :T_ARRAY]=>[2, 0, 0, 0, 0, 0]}}
{:kwargs_2=>{["kwargs_allocation.rb", 41, :T_ARRAY]=>[1, 0, 0, 0, 0, 0]}}

=======

Other questions will follow

Resources

Blog post, gist, google doc :

Tools:

Learn with PR comments and ruby issues:

Book :

Videos:

Youtube video playlist : https://www.youtube.com/playlist?list=PLXvaGTBVk36uIVBGKI72vqd9BFcMmPFI7

Audio:

Other than youtube :

Tanks 💕

And I would love to thanks especially Richard Schneems, Aaron Patterson, Sam Saffron, Nate Berkopec, Koichi Sasada...

About

Trying to understand Ruby memory 🌝

Resources

Readme
Activity

Stars

1 star

Watchers

2 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages

  • Ruby 100.0%