Just zipping up some photos.. this leaks memory. How can I optimize this routine ?

Downloading in Ruby Zipping Shelled Out.rb

photos.each do |photo|
  path = photo.photo.path
  file_path = zip_path + "/" + path
  FileUtils.mkdir_p(File.dirname(file_path))
  File.open(file_path, 'wb') do |file|
    file.write photo.photo.file.read
    file.close # shouldn't make a difference
  end
  # FWIW, this doesn't make a difference
  # file = File.open(file_path, 'wb')
  # file.write photo.photo.file.read
  # file.close
end

system("cd #{temp_dir} && zip #{zip_path} -r #{zip_name}")

Downloading Zipping Shelled Out.rb

# This seem sot be my solution of avoiding a huge amount of memory usage. (Thanks to [hanke](http://twitter.com/hanke) & [alloy](http://twitter.com/alloy) )

FileUtils.mkdir_p(zip_path)

photos.each do |photo|
  url = photo.photo_url
  path = photos.photo.path
  system("cd #{zip_path} && curl -L #{url} -o #{path} --create-dirs")
end

system("cd #{temp_dir} && zip #{zip_path} -r #{zip_name}")

Zipping Downloading in Ruby.rb

# This was my initial attempt which was altered transformed to what can be seen in Downloading in Ruby Zipping Shelled Out.rb
Zip::ZipOutputStream.open(zip_path) do |zos|
  photos.each do |photo|
    zos.put_next_entry(photo.photo.path)
    zos.write photo.photo.file.read
  end
end

• According to the docs I should not have to call "file.close" inside the File.open block since it should close the file automatically.
• Shelled Out.rb still increases memory usage until an unworkable high level (500 MB which is very close the the size of the images being loaded).
• When inspecting the process while running with Activity Monitor no more then 1 file is open at a time. So my assumption is this is not the problem.
• The memory increases before the zipping ever takes place so this isn't the problem either.
• All associations of photo are eagerly loaded which should therefor not be the problem either (and the DB is no where near 500 MB)
• The memory pain is obviously in the photo.photo.file.read call.
• The photo.photo.file.read call comes from this file in CarrierWave Fog
• Turns out the photo.photo.file.read call caches every single file read. Which prevents GC.

Just a note to the last line here: It only prevents GC'ing of the downloaded file body because there's a reference chain to the file. A clear method would be nice, or similar.