fix: avoid protobuf debug reflection in native images

[arnabnandy7]

Summary

Fixes #1995.

This PR avoids GraalVM native-image failures when using the protobuf debug exposition format with native histograms. The debug output path now formats protobuf metric families through direct generated protobuf getters instead of TextFormat.printer(), which relied on reflection unavailable in native images.

Changes

• Added a reflection-free protobuf debug formatter.
• Updated PrometheusProtobufWriterImpl#toDebugString(...) to use the new formatter.
• Added a native histogram debug-string regression test.

Testing

• mise run lint:fix
• ./mvnw.cmd -pl prometheus-metrics-exposition-formats test -Dcoverage.skip=true -Dcheckstyle.skip=true -Dwarnings=-nowarn
• ./mvnw.cmd -pl prometheus-metrics-exposition-formats-shaded -am install -DskipTests -Dcoverage.skip=true

[jaydeluca]

how do we know this fixes the graal specific issue? were you able to reproduce it?

your PR description also says "Note: Docker/Testcontainers exporter integration test was not run locally because Docker Desktop’s Linux engine was not running." which i think is just something you and your local llm need to sort out?

[arnabnandy7]

how do we know this fixes the graal specific issue? were you able to reproduce it?

your PR description also says "Note: Docker/Testcontainers exporter integration test was not run locally because Docker Desktop’s Linux engine was not running." which i think is just something you and your local llm need to sort out?

I'll try to find a solution to this and give you a reproducable steps or results.

[arnabnandy7]

@jaydeluca I reproduced the GraalVM native-image debug protobuf issue on Windows without Docker.

Environment:

• Windows
• Liberica NIK / GraalVM Native Image JDK 21
• Visual Studio 2022 Build Tools x64 compiler
• Reproducer: native-debug-repro

Important: run mvn clean package in the repro project between branches, otherwise the shaded repro jar can be stale.

Before, on main:

1. cd client_java
2. git checkout main
3. .\mvnw.cmd clean install -pl prometheus-metrics-exposition-formats-shaded -am -Dmaven.test.skip=true -Dcoverage.skip=true -Dcheckstyle.skip=true -Dwarnings=-nowarn
4. cd C:\Users\Arnab\Downloads\native-debug-repro
5. mvn -q -DskipTests clean package
6. native-image -jar target\native-debug-repro.jar -o native-debug-repro-main --no-fallback -H:+ReportExceptionStackTraces
7. .\native-debug-repro-main.exe

Result on main:

=== FAILURE: exception while formatting debug string ===
java.lang.IllegalStateException: Generated message class "...Metrics$MetricFamily" missing method "getName".
...
at io.prometheus.metrics.shaded.com_google_protobuf_4_35_0.TextFormat$Printer.printToString(...)
at io.prometheus.metrics.expositionformats.internal.PrometheusProtobufWriterImpl.toDebugString(...)
Caused by: java.lang.NoSuchMethodException: ...Metrics$MetricFamily.getName()

After, on fix/graalvm-native-histogram:

1. cd client_java
2. git checkout fix/graalvm-native-histogram
3. .\mvnw.cmd clean install -pl prometheus-metrics-exposition-formats-shaded -am -Dmaven.test.skip=true -Dcoverage.skip=true -Dcheckstyle.skip=true -Dwarnings=-nowarn
4. cd C:\Users\Arnab\Downloads\native-debug-repro
5. mvn -q -DskipTests clean package
6. native-image -jar target\native-debug-repro.jar -o native-debug-repro-fix --no-fallback -H:+ReportExceptionStackTraces
7. .\native-debug-repro-fix.exe

Result on fix branch:

=== debug=prometheus-protobuf output ===
name: "request_latency_seconds"
help: "request latency in seconds"
type: HISTOGRAM
...
=== SUCCESS: no exception thrown ===

Attaching the tool(native-debug-repro.zip) I used for the above test.

Seeking your re-review.

[zeitlinger]

https://github.com/prometheus/client_java/blob/main/.github/workflows/native-tests.yml run graalvm tests which runs https://github.com/prometheus/client_java/blob/main/integration-tests/it-spring-boot-smoke-test/src/test/java/io/prometheus/metrics/it/springboot/ApplicationTest.java using graalvm - can you add a reproducer there?

[arnabnandy7]

can you add a reproducer there?

@zeitlinger I have added a reproducer to the existing Spring Boot native smoke test in ApplicationTest. It now exercises PrometheusProtobufWriterImpl.toDebugString() with a native histogram snapshot, which is the code path that previously delegated to protobuf TextFormat reflection and failed under GraalVM native image. This should make the existing native-tests.yml job catch the regression directly.

PS: The latest failure looks unrelated to the PR changes, you probably need to create the api-change label in the repo which is why it's failing.

[zeitlinger]

@arnabnandy7 I removed arnabnandy7@c9c70bb#diff-9c5fb3d1b7e3b0f54bc5c4182965c4fe1f9023d449017cece3005d3f90e8e4d8 - as the ctor is not taking a shaded class any more - seems to work

[zeitlinger]

thanks for the contribution 😄

[arnabnandy7]

thanks for the contribution 😄

Thanks for clarifying and cleaning things up for me.