[GraalVM] v1.58.2 breaks native image builds - profiler ClassNotFoundException and SIGBUS crash

[dbernar1]

Tracer Version(s)

1.58.2

Java Version(s)

21

JVM Vendor

Oracle GraalVM

Bug Report

Summary

dd-trace-java v1.58.2 breaks GraalVM native image builds. v1.58.1 works correctly. The issue appears to be caused by the "Bump java profiler to 1.36.1" change in v1.58.2.

Environment

• dd-trace-java version: v1.58.2 (broken), v1.58.1 (works)
• GraalVM version: GraalVM 21.0.5+9.1
• Quarkus version: 3.28.4
• OS: macOS (also reproduced in Linux CI environment)

Steps to Reproduce

1. Configure Quarkus native build with dd-java-agent attached via -J-javaagent:

   quarkus.native.additional-build-args=\
     -J-javaagent:${AGENT_PATH:/path/to/dd-java-agent.jar},\
     -J-Ddd.logs.injection=true

2. Build native image with dd-java-agent v1.58.2
3. Build fails

Error Output

Stage 1 - ClassNotFoundException:

[dd.trace ...] ERROR datadog.common.version.VersionInfo - No version file found
Error: Class initialization of com.datadoghq.profiler.ThreadContext failed.
Caused by: java.lang.ClassNotFoundException: com.datadoghq.profiler.BufferWriter9

Stage 2 - After adding --initialize-at-run-time=com.datadoghq.profiler.ThreadContext:

The build progresses further but then crashes with a SIGBUS signal:

# A fatal error has been detected by the Java Runtime Environment:
#  SIGBUS (0xa) at pc=0x..., pid=..., tid=...
Error: Image build request for 'application-runner' ... failed with exit status 134

Version Comparison

Agent Version	Release Date	Native Build
v1.57.0	Dec 18, 2025	✅ SUCCESS
v1.58.1	Jan 21, 2026	✅ SUCCESS
v1.58.2	Jan 23, 2026	❌ FAIL

Root Cause Analysis

The v1.58.2 release notes mention "Bump java profiler to 1.36.1". This profiler update appears to have introduced classes or initialization patterns incompatible with GraalVM's native-image static analysis.

Workaround

Pin to v1.58.1 or earlier until this is fixed.

Expected Behavior

See bug report. Expecting it to not fail.

Reproduction Code

No response

[dbernar1]

FYI, I don't have past experience working on this part of the stack, so I would not have tried to reproduce or pin this kind of thing down without AI. This is just AI output of what it thought was the problem based on my prompts.

[jbachorik]

Hi @dbernar1 - thanks for the report, will take a look ASAP.