ddtrace breaks Dir.glob on gcsfuse filesystems

[dbuckley]

Current behaviour

irb(main):001:0> Dir.ancestors
=> [Dir, Enumerable, ActiveSupport::Dependencies::ZeitwerkIntegration::RequireDependency, Datadog::Profiling::Ext::Forking::Kernel, ActiveSupport::ForkTracker::CoreExtPrivate, ActiveSupport::ForkTracker::CoreExt, ActiveSupport::ToJsonWithActiveSupportEncoder, Object, Delayed::MessageSending, PP::ObjectMixin, JSON::Ext::Generator::GeneratorMethods::Object, ActiveSupport::Dependencies::Loadable, ActiveSupport::Tryable, Kernel, BasicObject]
irb(main):002:0> Dir.glob('/gcsfuse/t*')
=> []

Expected behaviour

irb(main):001:0> Dir.ancestors
=> [Dir, Enumerable, ActiveSupport::Dependencies::ZeitwerkIntegration::RequireDependency, ActiveSupport::ForkTracker::CoreExtPrivate, ActiveSupport::ForkTracker::CoreExt, ActiveSupport::ToJsonWithActiveSupportEncoder, Object, Delayed::MessageSending, PP::ObjectMixin, JSON::Ext::Generator::GeneratorMethods::Object, ActiveSupport::Dependencies::Loadable, ActiveSupport::Tryable, Kernel, BasicObject]
irb(main):002:0> Dir.glob('/gcsfuse/t*')
=> ["/gcsfuse/test.html", "/gcsfuse/test.txt"]

Steps to reproduce
Mount bucket via gcsfuse, load rails with ddtrace, call Dir.glob on any path under the gcsfuse mount point (other fs work as expected).

Environment

• ddtrace version: 1.20.0
• Configuration block (Datadog.configure ...):

Datadog.configure do |c|
 if (service_name = ENV["DD_SERVICE"])
   c.tracing.instrument :rails, service_name: service_name
   c.env = ENV.fetch("DD_ENV")
 end
end

• Ruby version: 2.7.7
• Operating system: Debian 11
• Relevant library versions: Rails 6.1.7.6

[ivoanjo]

Hey! 👋

This is a really interesting issue -- I wouldn't expect ddtrace to have any impact on globbing; we don't particularly monkey patch anything related to accessing files.

Can you reproduce this outside a Rails app? And are you enabling any other features (profiling, asm, etc) via environment variables, or are you only setting up distributed tracing?

[ivoanjo]

Oh, I just spotted there's a support ticket about this. Let's focus the debugging there, so I'm going to go ahead and close this issue :)

[ivoanjo]

As a record, here's some details about this:

• This is caused by a bug in the Ruby VM handling of signal interruptions
• The workaround from https://docs.datadoghq.com/profiler/profiler_troubleshooting/ruby/#unexpected-failures-or-errors-from-ruby-gems-that-use-native-extensions-in-dd-trace-rb-1110 works to tackle this issue

We're also preparing a better fix that will be applied automatically.

[TonyCTHsu]

👋 @dbuckley We just released v2.2.0. Give it a try!