Auto‑Instrumenting Java in GraalVM Native Images: A Practical OTel Solution

Background

OpenTelemetry's influence is growing; Alibaba Cloud contributes heavily to OTel Java Instrumentation, ranking first in APAC. At the 2024 OpenTelemetry Community Day, engineers presented a solution for automatically enhancing OTel Java Agent under GraalVM static compilation.

Challenges

Serverless Java apps suffer long startup times and high memory usage due to JVM initialization, class loading, and JIT compilation. GraalVM Native Image removes the VM, reducing startup time and memory, but disables dynamic features like reflection and Java Agent bytecode rewriting.

Solution Overview

The proposed approach introduces a static Java Agent instrumentation that works before compilation. It consists of a pre‑run phase where OTel Java Agent and a Native Image agent collect transformed classes, and a static compilation phase that replaces original classes with the transformed ones, producing an executable with built‑in observability.

Implementation Details

Native Image agent is extended with an interceptor to record class bytecode changes. Additional adaptations handle JDK classes, classloader differences, and shading. Class replacement can be done via the JDK -classpath or --module‑path patching mechanisms.

Demo & Results

Videos demonstrate pre‑execution, static compilation, and runtime of the compiled binary, showing startup time reduced to ~0.07 s and memory usage dramatically lower while still collecting traces and metrics. Benchmarks on Spring Boot, Kafka, Redis, and MySQL confirm the benefits.

Future Plans

Further work includes comprehensive testing of metrics, traces, and logs, and merging the pre‑run phase into compilation to capture all transformed classes, reducing reliance on test coverage.

Community Involvement

Alibaba Cloud continues to contribute to OpenTelemetry and GraalVM communities, with PRs, issue discussions, and upcoming open‑source releases, and will present at KubeCon + CloudNativeCon 2024.