wicked_pdf: Thread puma threadpool terminated with exception machine stack overflow in critical region (fatal)

Issue description

Our CI server is having a problem with the recent major bump to 2.0. Locally we are fine and pdf’s work. If I downgrade to 1.4.0 our CI is fine.

Seems this was also an issue in the past: https://stackoverflow.com/questions/55280153/machine-stack-overflow-in-critical-region-fatal https://github.com/mileszs/wicked_pdf/issues/810

Randomized with seed 34635
Capybara starting Puma...
* Version 4.3.3 , codename: Mysterious Traveller
* Min threads: 0, max threads: 4
* Listening on tcp://127.0.0.1:34469
#<Thread:0x0000561dc0601600@puma threadpool 001@/home/circleci/project/vendor/bundle/ruby/2.6.0/gems/puma-4.3.3/lib/puma/thread_pool.rb:89 run> terminated with exception (report_on_exception is true):
Traceback (most recent call last):
/home/circleci/project/vendor/bundle/ruby/2.6.0/gems/rack-2.2.2/lib/rack/urlmap.rb:58:in `each': machine stack overflow in critical region (fatal)

https://github.com/puma/puma/blob/master/lib/puma/thread_pool.rb#L89 https://github.com/rack/rack/blob/master/lib/rack/urlmap.rb#L58

Expected or desired behavior

CI runs and passes

System specifications

wicked_pdf 2.0.1 wkhtmltopdf-binary-edge 0.12.4.0 because of https://github.com/pallymore/wkhtmltopdf-binary-edge/issues/15 and extra build config just to get it to work so we locked it Ubuntu 16.04 ruby 2.6.5 rails 6.0.2.1

About this issue

Original URL
State: open
Created 4 years ago
Reactions: 1
Comments: 23 (7 by maintainers)

Most upvoted comments

@Jeremie1707 I tried to quickly setup an app to reproduce this, and was not able to. I looked through the commits of rails_admin and see they recently made some updates to improve compatibility with the zeitwerk loader, so maybe it’s better now? You could try upgrading rails_admin and remotipart with:

bundle update rails_admin remotipart

And try again.

Another thing that might help and be minimally invasive is to defer requiring wicked_pdf until railties are booted, and your app initializers run. You can do this by editing Gemfile like this:

gem 'wicked_pdf', require: false

Then requiring wicked_pdf before it’s needed in the initializer:

require 'wicked_pdf'
WickedPdf.config = {
  exe_path: '/usr/local/bin/wkhtmltopdf'
}

Then only in the controllers it’s needed, add the PDF helpers:

class MyController < ApplicationController
  prepend WickedPdf::PdfHelper
  include WickedPdf::WickedPdfHelper::Assets

  def show
  end
end

Or, you can use @darrenterhune 's fork by just changing out your Gemfile

gem 'wicked_pdf', git: 'https://github.com/darrenterhune/wicked_pdf'

Please let me know what you end up doing and what worked or not.

unixmonkey on Mar 26, 2020

Ok @unixmonkey I had some more time this avo to debug this. I forked this gem and made this change https://github.com/darrenterhune/wicked_pdf/commit/ccc861596dcd8e43e5243a79b8658cf7c3c9b4e0 and it’s now working and passing CI. I just wanted to make sure this comment was true https://github.com/mileszs/wicked_pdf/pull/869/files#r386040878

I also found this where some people are saying having ActiveSupport.on_load ran in multiple gems causes them issues https://github.com/rails/rails/issues/27013 which might be a problem (but I have no idea why) because remotipart also does this https://github.com/JangoSteve/remotipart/blob/master/lib/remotipart/rails/railtie.rb#L33-L36

We just need to 🔪 rails_admin but we are still months off from that and I suspect others will also have this issue.

darrenterhune on Mar 10, 2020

This is no longer an issue since v2.6.0 was released back in May.

I’ve just removed this from my Gemfile:

# Tracking master to resolve: https://github.com/mileszs/wicked_pdf/issues/891
gem "wicked_pdf", github: "mileszs/wicked_pdf", branch: "master"

and upgraded to version v2.6.3.

davegudge on Jun 24, 2022

Issue appears resolved with master branch 5cee578699fe3e5015eb7d76b604a0c121266018

At least CI is happy, I have not test deployed but should be alright.

mathieujobin on Sep 18, 2021

@feliperaul the same thing really weird, thanks for suggestion

almokhtarbr on Feb 10, 2021

@almokhtarbr Nope, and I can’t even reproduce it, it’s totally random (every once 15 days for us).

For now I have switched to PDFKit, the APIs are very similar.

feliperaul on Feb 10, 2021

Does anyone in this thread can reliably reproduce this bug?

I’m also facing it, but only randomly.

Nate Berkopec, a Puma maintainer, is looking into this on https://github.com/puma/puma/issues/2552, but he also can’t reproduce the “fatal - machine stack overflow in critical region” error.

Please report in that issue if you can help.

feliperaul on Feb 9, 2021

Yep. Thanks for confirming that. I’m working on a replacement system that avoids any kind of conflict, but I suspect that won’t be released for a while either. You might be better off using your fork or monkey-patching the initializer as a no-op, and including/prepending in the controllers you need it in until then.

unixmonkey on Mar 10, 2020