Garbarge Collection in Ruby with Circular Object References

Question

I'm having an issue with garbage collection in Ruby where an object that I think should be garbage collection is not being garbage collected.

require 'ruby-mass'

def find_dependencies(_object_id,_mapped = {})
  mapped = _mapped
  points_to_object = Mass.references(Mass[_object_id])
  ids = points_to_object.keys.map{|x| /\#(\d*)/.match(x).captures.first.to_i}
  mapped[_object_id] = ids

  unmapped = ids - mapped.keys
  unmapped.each do |x|
    new_deps = find_dependencies(x,mapped)
    mapped.merge(new_deps)
  end
  mapped
end

Do some stuff that makes the objects, and find the relevant object ID. GC.start, then:

> find_dependencies(144789180)
=> {144789180=>[61895480, 144786340, 147807540],
 61895480=>[144789180],
 144786340=>[144789180],
 147807540=>[144789180]}

It looks like there is a circular reference pattern here, but it is all completely contained in these four objects, so the Mark-and-Sweep collector should find them and remove them.

So, either there is a bug in my find_dependencies_function, the Mass gem, or Ruby's garbage collector. How do I narrow this down to find out what the problem is and solve this memory leak?

Answer 1

The GC in Ruby works essentially like so:

1. Mark all global objects as live.

2. Sweep through objects, garbage collect unless a parent is live.

So, in the case of a circular reference, A holding onto B holding onto A would get GC'd because neither is being held by a live object.

Per the comments, something is necessarily holding onto the object somewhere... Or maybe Mass is catching RangeError or something...

>> a = {}
=> {}
>> a[:a] = a
=> {:a=>{...}}
>> a.object_id
=> 2269556540
>> a = nil
=> nil
>> GC.start
=> nil
>> ObjectSpace._id2ref(2269556540)
RangeError: 0x8746af3c is recycled object
    from (irb):17:in `_id2ref'
    from (irb):17