Large Dictionaries Not Released From Memory - Python + Ubuntu

13 Oct 2014

I have a flask application running on Ubuntu which reads from a file and creates several large dictionaries. After the dictionaries are created and flask returns the request, the memory used for the dictionaries is not released back to the OS. This example shows the problem:

import random import optparse import gc from flask import redirect, Flask from memory_profiler import memory_usage # Create flask app app = Flask(__name__) app.debug = True @app.route('/') def hello_world(): ### # The test: # Generate a 2440 mb dictionary and return a blank page. ### large_dict = {} for x in xrange(125000): large_row = {} for y in xrange(125): large_row[random.randint(1, 10000000000)] = random.randint(1, 10000000000) large_dict[random.randint(1, 10000000000)] = large_row large_dict = {} gc.collect() return '' @app.after_request def after(response): print memory_usage(-1, interval=.2, timeout=1), "after request" return response def main(): app.config['SECRET_KEY'] = "BLAH" parser = optparse.OptionParser() parser.add_option("--host", default="0.0.0.0") # can't use -h, it's already taken by help parser.add_option("-p", "--port", default="5123") options, args = parser.parse_args() if len(args) > 0: parser.error("Unexpected arguments: %s" % args) app.run(host=options.host, port=int(options.port)) if __name__ == '__main__': # Start app main()

It’s not really a Flask issue, but the Flask guys were super helpful when I asked why my large dictionary wasn’t released from memory: Flask issue #1202

I dug even deeper and learned this is a Linux thing:
“Python returns memory to the OS on the heap (that allocates other objects than small objects) only on Windows, if you run on Linux, you can only see the total memory used by your program increase.”
Theano: Python memory management

My solution:
Turning it into a function and running it as a separate process:

import random import optparse import gc from flask import redirect, Flask from memory_profiler import memory_usage from multiprocessing import Process # Create flask app app = Flask(__name__) app.debug = True @app.route('/') def hello_world(): ### # The test: # Generate a 2440 mb dictionary and return a blank page. ### def separate_process(): large_dict = {} for x in xrange(125000): large_row = {} for y in xrange(125): large_row[random.randint(1, 10000000000)] = random.randint(1, 10000000000) large_dict[random.randint(1, 10000000000)] = large_row p = Process(target=separate_process) p.start() p.join() # this blocks until the process terminates return '' @app.after_request def after(response): print memory_usage(-1, interval=.2, timeout=1), "after request" return response def main(): app.config['SECRET_KEY'] = "BLAH" parser = optparse.OptionParser() parser.add_option("--host", default="0.0.0.0") # can't use -h, it's already taken by help parser.add_option("-p", "--port", default="5123") options, args = parser.parse_args() if len(args) > 0: parser.error("Unexpected arguments: %s" % args) app.run(host=options.host, port=int(options.port)) if __name__ == '__main__': # Start app main()

If you want to go even deeper down the rabbit hole, read these:

“most malloc implementations will not release memory to the operating system, and the few that do, do not do it very easily” -http://stackoverflow.com/questions/2215259/will-malloc-implementations-return-free-ed-memory-back-to-the-system
“We compiled a version of Python with TCMalloc that only uses mmap. When testing the new Python in one of our largest projects, we found that not only did Python give back memory to the OS correctly, it also had a reduced memory usage and no apparent CPU penalty for using mmap instead of brk.” - pushingtheweb.com: Python and TCMalloc