Sep 6, 2016

Ruby is Dead! (You nead to take care of its memory issues)

One of the common problems with Ruby is its memory usage, just like JVM based languages (yes Java, I'm talking about you). Actually Ruby just like Java is based on garbage collector. This GC if not used correctly, can cause "stop the world scenarios" (as we can see in the attached diagram). In this cases, GC actually stops responding while consuming servers whole CPU and in some cases even causes server reboot.
Unicorn and Ruby memory leak causes server downtime
What can be done?

1. Reboot your Ruby Periodically

If you are using the popular Unicorn web server, memory issues may be even more severe, as Unicorn forks. During forking, it's copying the whole memory footprint of the parent (Copy on Write or CoW). Therefore, you may use the "Unicorn Worker Killer" gem that will monitor your server and gracefully will restart specific worker when memory reaches a new high or specific number of requrests were served. Since the gem supports randomization, it assures in a very high probablilty that the server will continue server.

Installing the killer:
gem 'unicorn-worker-killer'

Add to above "require ::File.expand_path('../config/environment', __FILE__)"
# Unicorn self-process killer
require 'unicorn/worker_killer'

And decide your worker graceful reboot method: 
# Max requests per worker
use Unicorn::WorkerKiller::MaxRequests, 3072, 4096
# Max memory size (RSS) per worker
use Unicorn::WorkerKiller::Oom, (192*(1024**2)), (256*(1024**2))
If you are using Ruby 2.X, you can exploit the CoW by better configuring Unicorn.
config/unicorn.rb (and a detailed explenation on setting at sirupsen)
 - worker_processes: 1x your cores
 - timeout: worker request timeout, should be 15 to 30 sec max
 - preload_app: enables CoW, but requires managning connect/disconnect on fork

3. Take care of GC configuration
You can take a look at a detailed discussion at collective idea's post.

Bottom Line
Dynamic languages has their downs, yet w/ the right design you can keep them up and running,

Keep Performing,


Intense Debate Comments

Ratings and Recommendations