Over a million developers have joined DZone.
{{announcement.body}}
{{announcement.title}}

The Fastest Way to Generate a Heap Dump on Heroku

DZone's Guide to

The Fastest Way to Generate a Heap Dump on Heroku

Using Ruby on Rails, we go over how to do what a lot of devs consider to be a last resort, generating a hope dump to open up more memory in you application.

· Web Dev Zone ·
Free Resource

Learn how error monitoring with Sentry closes the gap between the product team and your customers. With Sentry, you can focus on what you do best: building and scaling software that makes your users’ lives better.

You’ve got an app with runaway memory use, what do you do hotshot? What do you do? If you’ve exhausted the usual suspects it might be time to take drastic steps. It might be time to take a production heap dump. I previously wrote about doing this on Heroku, but since then we’ve launched Heroku exec, a way to SSH into a live running Dyno to allow you to debug. Now that you can do that, you don’t need an AWS account or any fancy gems to generate a heap dump, just activate this feature and add the rbtrace gem to your app. Let’s do this to an app together.

First, we need to set up the app with Heroku Exec. Check the Heroku Exec docs as these steps may change in the future. We’re going to start by installing the plugin:

$ heroku plugins:install heroku-cli-exec

Next, execute:

$  heroku ps:exec
Creating heroku-exec:test on issuetriage... free
Adding the Heroku Exec buildpack to issuetriage

Run the following commands to redeploy your app, then Heroku Exec will be ready to use:
  git commit -m "Heroku Exec initialization" --allow-empty
  git push heroku master

This will add a free add-on to your app to allow you connect to your Dynos. But to be able to use it, you have to deploy. You can do this by running the following:

$ git commit -m "Heroku Exec" --allow-empty
$ git push heroku master

Once your app is done deploying, you can SSH into a running Dyno by executing:

$ heroku ps:exec
Establishing credentials... done
Connecting to web.1 on ⬢ issuetriage...
~ $

Before we can take a heap dump, we’ll need to tell your app to start tracing object allocations. Get out of your ps:exec session if you haven’t already.

Add the rbtrace gem to your Gemfile:

# Gemfile

gem "rbtrace"

Then run bundle install locally. Next, you’ll want to tell your app to start tracing allocations. If you’re using a forked webserver, like Unicorn or Puma, you can enable this in an on_worker_boot block, or something similar.

If you’re using Puma, add this code to your config/puma.rb file:

on_worker_boot do
  if ENV['DYNO'] == 'web.1'
    # Trace objects on 1 Dyno to generate heap dump in production
    require 'objspace'
    ObjectSpace.trace_object_allocations_start
  end
end

Here we’re telling any worker processes to trace where objects were created at. This is not “free” and will cause the Dyno to run slightly slower, because of this we want to limit to running this code in just one Dyno, web.1.

You’ll want to commit this and deploy it. Once you’ve done that you can now SSH into your Dyno and generate the Heap dump.

$ heroku ps:exec
Establishing credentials... done
Connecting to web.1 on ⬢ issuetriage...
~ $

First, we’ll need to figure out the PID of your worker processes.

~ $ ps -eo pid,comm,ppid | grep ruby
   29 ruby                3
   51 ruby               29

We’re looking at two Ruby processes in our Dyno. The first has a PID of 29 and a parent PID of 3. The second has a PID of 51 and a parent PID of 29. This means that 51’s parent is also a Ruby process. Puma uses a master process to hand off requests to child or “worker” processes. We want to run our heap dump against our “child process,” which is 51.

Note your PID will be different, use the same technique to determine your child Ruby processes.

We can generate the heap dump using rbtrace with this command:

$ bundle exec rbtrace -p 51 -e 'Thread.new{GC.start;require "objspace";io=File.open("/tmp/ruby-heap.dump", "w"); ObjectSpace.dump_all(output: io); io.close}'

This will generate a heap dump in the file /tmp/ruby-heap.dump.

We can now download this file to our local machines. Exit the ps:exec sessions. The file continues to live on the Dyno because the Dyno has not restarted yet. You can now run this command:

$ heroku ps:copy /tmp/ruby-heap.dump

This will download the file and put it in your local directory with the name ruby-heap.dump.

We can now use debugging tools on this dump file. Such as the heapy gem that I wrote and maintain.

$ heapy read ruby-heap.dump

Analyzing Heap
==============
Generation: nil object count: 302087, mem: 0.0 kb
Generation:   4 object count: 171, mem: 6174.9 kb
Generation:   5 object count: 4769, mem: 1614.3 kb
Generation:   6 object count: 1, mem: 0.2 kb
Generation:   8 object count: 199, mem: 8.6 kb
Generation:   9 object count: 1, mem: 0.0 kb
Generation:  12 object count: 44, mem: 198.3 kb
Generation:  13 object count: 6, mem: 0.5 kb
Generation:  14 object count: 645, mem: 1079.0 kb
Generation:  15 object count: 74, mem: 5.1 kb

Heap total
==============
Generations (active): 10
Count: 307997
Memory: 9080.9 kb

You can dig into a generation using the following code:

$ heapy read ruby-heap.dump 5

Analyzing Heap (Generation: 5)
-------------------------------

allocated by memory (1653008) (in bytes)
==============================
  1049992  /app/vendor/bundle/ruby/2.4.0/gems/rack-timeout-0.4.2/lib/rack/timeout/support/scheduler.rb:67
   170671  /app/vendor/bundle/ruby/2.4.0/gems/json-2.0.3/lib/json/common.rb:224
    88623  /app/vendor/bundle/ruby/2.4.0/gems/multi_json-1.12.1/lib/multi_json/adapters/oj.rb:15
# ...

Here we can see the three locations that allocated the most memory.

I’ll admit that I’ve never successfully found an application level memory problem from a heap dump, but I know some have. If you’re out of other options, it’s another thing you can try. The exciting thing about ps:exec is that you’re not limited to heap dumps, you now have access to all system level debugging tools for live running processes. For instance, you could install gdb or lsof via the Heroku APT buildpack to debug in a way you never could before. Keep in mind that if you’re running on a regular Dyno, it’s in a container and on a shared system with other applications so using something like ps to get memory use won’t be entirely accurate.

I’m excited for the power and flexibility that ps:exec brings to Heroku.

What’s the best way to boost the efficiency of your product team and ship with confidence? Check out this ebook to learn how Sentry's real-time error monitoring helps developers stay in their workflow to fix bugs before the user even knows there’s a problem.

Topics:
web dev ,ruby ,heap dump

Published at DZone with permission of

Opinions expressed by DZone contributors are their own.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}