[VL] Spark 4.1 CI intermittently aborts during Velox backend initialization

[zml1206]

Backend

VL (Velox)

Bug description

The spark-test-spark41 workflow intermittently fails while running unit tests in the backends-velox module.
Scala compilation completes successfully. The failure occurs after Surefire starts the forked JVM and initializes the Velox backend. Both libgluten.so and libvelox.so are loaded successfully, and the Velox components are registered. The JVM then aborts unexpectedly with exit code 134 (SIGABRT) before the native-dependent test starts executing.

The forked VM terminated without properly saying goodbye.
Process Exit Code: 134
Corrupted channel by directly writing to native stream in forked JVM 1

https://github.com/apache/gluten/actions/runs/29471701089/job/87536355426?pr=12530
https://github.com/apache/gluten/actions/runs/29478222210/job/87556349046

Gluten version

No response

Spark version

None

Spark configurations

No response

System information

No response

Relevant logs

[PerumalsamyR]

This appears to have become consistent rather than intermittent. spark-test-spark41 and spark-test-spark41-slow are failing with the same signature (exit 134 / SIGABRT in the backends-velox surefire fork during ColumnarBatchTest, right after backend initialization, with "Corrupted channel by directly writing to native stream in forked JVM 1") on every recent Velox Backend (x86) run I checked today:

• https://github.com/apache/gluten/actions/runs/29583464547 (PR [GLUTEN-12238][VL] Translate Velox invalid bitmap position errors to Spark's INVALID_BITMAP_POSITION exception #12544, both jobs, twice in a row after a retrigger)
• https://github.com/apache/gluten/actions/runs/29573801822
• https://github.com/apache/gluten/actions/runs/29572888232
• https://github.com/apache/gluten/actions/runs/29572334274

All fail before the gluten-ut modules run, so Spark 4.1 UT coverage is currently blocked repo-wide.

[FelixYBW]

We will take a look now