Inspector.targetCrashed received when `no_sandbox: true` on Chromium 115 on Linux about chromic_pdf

[maltoe]

Hey @aptinio,

can you share some details on your setup? Chromium version that has the error, but also host OS, environment, Elixir version, etc.. So we can try to reproduce this error.

[aptinio]

This happens on my dev environment:

• Arch Linux
• asdf-managed Elixir and Erlang/OTP versions:
  • 1.15.4-otp-25 and 25.3.2.5
  • 1.15.4-otp-26 and 26.0.2
• Chromium 115.0.5790.170

and on prod too:

• Docker image hexpm/elixir:1.15.4-erlang-26.0.2-debian-bookworm-20230612-slim
• same Chromium version 115.0.5790.170

My chromic_pdf_options:

  defp chromic_pdf_options do
    [
      disable_scripts: true,
      no_sandbox: true,
      offline: true
    ]
  end

Thanks @maltoe!

[aptinio]

For now, I did this:

   defp chromic_pdf_options do
     [
       disable_scripts: true,
-      no_sandbox: true,
+      chrome_args: "--no-sandbox",
       offline: true
     ]
   end

[maltoe]

@aptinio Thanks for the report. I've been able to reproduce this bug on Debian Bookworm. I have not attempted to debug the bug / changed behaviour further within chrome or the more recent environment, and also have not found any matching issue in their trackers.

At the same time, I can't recall why the --no-zygote flag is present there at all. Puppeteer doesn't seem to have equivalent code, and the docs I could find just say that when you use --no-zygote, you must use --no-sandbox as well (but not vice versa). So, I think --no-zygote is a mistake here and am tempted to simple go for #272, though breaking existing setups is a concern of cause. I did check on Alpine 3.16 and Debian Buster that --no-zygote isn't needed to run an unsandboxed root chrome, so it may be fine.

@andreasknoepfle wdyt? Remove the flag and risk breaking people's setups (certainly change people's setup though, as removing --no-zygote will add the zygote process) or play it safe?

[maltoe]

"Playing it safe" could be: Deprecate --no-sandbox with a warning and change the docs to just tell people to use chrome_args as in @aptinio 's snippet.

[maltoe]

Decided to YOLO it this time as tests in our CI environments suggest that --no-zygote was never needed and hence it seems unlikely that anyone will notice it.

@aptinio If you can, if would be nice if you could verify that current main fixes the bug in your environments.

[aptinio]

It does. Thanks @maltoe! ☺️