MONGOID-5820 Fix the session object leak when using transactions

[Virus-X]

The issue

I found the memory leak inside Threaded class, when saving documents within transaction: session object is retained forever inside thread-local variable.

Affected version: Mongoid 9.0.2 / Ruby 3.3.4

Explanation

• Mongoid uses thread-local variables to temporarily hold modified_documents until transaction is committed.
• The [mongoid]:modified-documents contains a hash with session as key and Set of documents as value.
• During transaction commit phase, the Mongoid::Threaded::clear_modified_documents is called, which clears documents from the set, however, the hash inside [mongoid]:modified-documents still contains a session object as a key.
• This session object is then persists in memory, even after the session was finished by the Mongoid::Threaded::clear_session call.

Testing

I used the following code to reproduce the issue:

require_relative 'spec_helper'
require 'memory_profiler'

describe "Mongoid test" do
  let(:model) do
    Class.new do
      include Mongoid::Document

      store_in collection: 'tests'
      field :name, type: String
    end
  end

  context 'memory leak test' do
    it 'store in transaction' do
      run_memory_profiling do
        model.transaction do
          model.create!(name: 'test')
        end
      end
    end
  end

  def run_memory_profiling(&block)
    # Warmup
    yield

    report = MemoryProfiler.report do
      2000.times(&block)
      GC.start(full_mark: true, immediate_sweep: true)
    end
    p "Total retained: #{report.total_retained_memsize / 1024} KB (#{report.total_retained} objects). "
  end
end

Initial results:

Total retained: 3430 KB (36053 objects). 

Report shows a number of retained objects, which corresponds to 2000 iterations of test:

Results after modification:

Total retained: 6 KB (61 objects).

[jamis]

Looks good! Just one question about the commented-out bit in lite_spec_helper.rb, but I've got no material concerns.

[jamis]

Are these lines commented out because they might come back? Should we just delete them and rely on git to remember them if we need to resurrect them?

[comandeo-mongo]

Oh, good catch! This should be removed eventually, I believe, but not in scope of this PR. I'll revert.

[comandeo-mongo]

Hi @Virus-X , thank you very much for your contribution! I am sorry it takes me so much time to merge it. The reason was that we needed a test for the change; and while adding the test I discovered one more related issue that was also fixed.

[Virus-X]

Thanks! Was glad to help!