Segmentation fault in TracePoint handler on Ruby 3.4.6 +YJIT (Sidekiq)

[luizdalcico]

Issue Description

We're experiencing recurring segmentation faults in production on our Sidekiq background worker. The crash occurs inside Ruby's <internal:trace_point> implementation, called from sentry-ruby's include_local_variables TracePoint handler.

Crash Dump

<internal:trace_point>:382: [BUG] Segmentation fault at 0x00000001c63e5400
ruby 3.4.6 (2025-09-16 revision dbd83256b1) +YJIT +PRISM [x86_64-linux]

-- Control frame information -----------------------------------------------
c:0109 p:0003 s:0725 e:000724 METHOD <internal:trace_point>:382
c:0108 p:0021 s:0721 e:000720 BLOCK  sentry-ruby-5.27.0/lib/sentry-ruby.rb:74 [FINISH]
c:0107 p:0197 s:0715 e:000714 BLOCK  /usr/local/lib/ruby/3.4.0/resolv.rb:570
c:0106 p:0012 s:0703 e:000702 BLOCK  /usr/local/lib/ruby/3.4.0/resolv.rb:1168 [FINISH]
c:0105 p:0027 s:0698 e:000697 METHOD <internal:array>:228
c:0104 p:0007 s:0692 e:000691 BLOCK  /usr/local/lib/ruby/3.4.0/resolv.rb:1166
c:0103 p:0027 s:0688 e:000687 METHOD <internal:array>:228
c:0102 p:0006 s:0682 e:000681 BLOCK  /usr/local/lib/ruby/3.4.0/resolv.rb:1165
c:0101 p:0027 s:0678 e:000677 METHOD <internal:array>:228
c:0100 p:0028 s:0672 e:000671 METHOD /usr/local/lib/ruby/3.4.0/resolv.rb:1163
c:0099 p:0030 s:0664 e:000663 METHOD /usr/local/lib/ruby/3.4.0/resolv.rb:540
c:0098 p:0008 s:0654 e:000653 METHOD /usr/local/lib/ruby/3.4.0/resolv.rb:523
c:0097 p:0022 s:0647 e:000646 METHOD /usr/local/lib/ruby/3.4.0/resolv.rb:414
SEGV received in SEGV handler
./entrypoint_worker.sh: line 29:   110 Aborted (core dumped) bundle exec sidekiq

What We Know

• The crash happens inside <internal:trace_point>:382, called from sentry-ruby.rb:74 — the TracePoint handler registered by include_local_variables = true
• resolv.rb (Ruby's DNS resolver) is in the call stack — it raises internal exceptions as part of normal flow control, which triggers the global TracePoint on :raise
• A double SEGV occurred (SEGV received in SEGV handler), so the C-level backtrace was not printed
• Ruby GC heap metrics and memory usage showed no anomalies before the crash
• The Sidekiq process was running concurrent workloads

Reproduction Steps

Run concurrent jobs in the corresponding versions that might eventually raise exceptions

Environment

• sentry-ruby: 5.27.0
• Ruby: 3.4.6 (2025-09-16 revision dbd83256b1) +YJIT +PRISM [x86_64-linux]
• Sidekiq: background worker with ~25 concurrent threads
• Platform: Linux

Sentry Configuration

Sentry.init do |config|
  config.include_local_variables = true
  config.traces_sample_rate = 1.0
  config.profiles_sample_rate = 1.0
end

Expected Behavior

Not to crash sidekiq server

Actual Behavior

Crashing server

Ruby Version

3.4.6 (2025-09-16 revision dbd83256b1)

SDK Version

5.27.0

Integration and Its Version

Sidekiq

Sentry Config

Sentry.init do |config|
  config.include_local_variables = true
  config.traces_sample_rate = 1.0
  config.profiles_sample_rate = 1.0
end

[linear-code]

RUBY-164

[sl0thentr0py]

thanks for the report @luizdalcico
we simply register the TracePoint hook to capture variables so this is technically an interaction within Ruby. Since you're using all the latest features (YJIT + prism) it might be worthwhile to get the YJIT people to look at it.

Overall, I think TracePoint with YJIT is a known source of both overhead and in worse cases, segfaults so I would recommend not using include_local_variables if you're on a YJIT build. Sorry I can't give you a better answer than that for now.