[VL] Native validation fails when Spark session timezone is GMT

[malinjawi]

Backend

VL (Velox)

Bug description

Expected behavior:

Spark 3.5 queries on Gluten main should be able to use the Velox backend even when the effective Spark session timezone is GMT. Gluten should either pass a native-compatible timezone string to Velox or normalize the Spark value before native validation.

Actual behavior:

Gluten forwards the effective Spark session timezone into native query config. When that value is GMT, native validation fails before execution starts:

Exception: VeloxUserError
Error Source: USER
Error Code: INVALID_ARGUMENT
Reason: session 'session_timezone' set with invalid value 'GMT'
Expression: tz::getTimeZoneID(*tz, false) != -1
Function: validateConfig
File: .../ep/build-velox/build/velox_ep/velox/core/QueryConfig.cpp
Line: 44

As a result, the query never reaches the actual Delta/Iceberg/Velox execution path. Native offload is rejected at validation time because of the session timezone value alone.

This was reproduced on a clean detached worktree without the local timezone normalization patch, after a fresh native rebuild using Gluten's pinned Velox dependency. The failure is not Delta-only; it also reproduces in a non-Delta Velox suite, which indicates this is a general Spark-to-native integration bug in the Velox backend path rather than a Delta-specific semantic issue.

Reproduction summary:

1. Create a clean detached worktree at f350a440dbfa3a3412f0aed48a89fdac0b16ad48.

2. Rebuild native dependencies from the clean checkout:

   ./dev/builddeps-veloxbe.sh --run_setup_script=OFF --build_arrow=OFF --build_tests=ON --spark_version=3.5

3. Run the backend tests from the same clean checkout:

   ./build/mvn \
     -s /path/to/dev/maven-public-settings.xml \
     -Dmaven.repo.local=/path/to/.run-scala-test-cache/m2 \
     -pl backends-velox \
     -am \
     -Pbackends-velox,delta,iceberg,spark-3.5 \
     -Dtest=OptimizedWritesSuite \
     -Dsurefire.failIfNoSpecifiedTests=false \
     test

Observed failures include:

• org.apache.spark.sql.delta.DeltaColumnDefaultsInsertSuite
• org.apache.gluten.execution.VeloxDeltaSuite
• org.apache.gluten.execution.VeloxTPCHMiscSuite

Example clean repro reports:

• backends-velox/target/surefire-reports/TEST-org.apache.spark.sql.delta.DeltaColumnDefaultsInsertSuite.xml
• backends-velox/target/surefire-reports/TEST-org.apache.gluten.execution.VeloxDeltaSuite.xml
• backends-velox/target/surefire-reports/TEST-org.apache.gluten.execution.VeloxTPCHMiscSuite.xml

Why this matters:

• It blocks native validation before query execution.
• It prevents meaningful verification of Delta/Iceberg behavior because suites fail on config setup first.
• It hides deeper runtime issues behind a config gate.
• It affects the current clean main build path.

Why this appears to be a Gluten-side issue:

• Spark 3.5 can accept and surface GMT-family timezone IDs as an effective session timezone.
• Gluten forwards that session timezone into native config.
• The current clean Gluten product path fails before execution on that value.
• The practical compatibility boundary between Spark session config and native Velox config is in Gluten.

Candidate fix:

Normalize Spark GMT / GMT+/-offset session timezone values to native-safe UTC forms at the Gluten/native boundary before native validation.

Gluten version

main branch

Spark version

Spark-3.5.x

Spark configurations

No explicit spark.sql.session.timeZone override was set in the reproducer.

The effective session timezone seen by native validation was GMT.

System information

./dev/info.sh: line 47: lscpu: command not found

Gluten Version: 1.7.0-SNAPSHOT
Commit: f350a440dbfa3a3412f0aed48a89fdac0b16ad48
CMake Version: 4.2.3
System: Darwin-25.3.0
Arch: arm64
CPU Name: 
C++ Compiler: /usr/bin/c++
C++ Compiler Version: 17.0.0.17000013
17.0.0.17000013
C Compiler: /usr/bin/cc
C Compiler Version: 17.0.0.17000013
17.0.0.17000013
CMake Prefix Path: /Applications/Xcode_16.4.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr;/opt/homebrew;/usr/local;/usr;/;/opt/homebrew;/usr/local;/usr/X11R6;/usr/pkg;/opt;/sw;/opt/local


### Relevant logs

```bash
Exception: VeloxUserError
Error Source: USER
Error Code: INVALID_ARGUMENT
Reason: session 'session_timezone' set with invalid value 'GMT'
Retriable: False
Expression: tz::getTimeZoneID(*tz, false) != -1
Function: validateConfig
File: /Users/malinjawi/Documents/GitHub/GlutenVelox/incubator-gluten-verify-clean/ep/build-velox/build/velox_ep/velox/core/QueryConfig.cpp
Line: 44
```

[malinjawi]

Hey @zhztheplayer @rui-mo I spotted this issue in my local dev. Could you please confirm if this is a valid issue? I have an idea for the fix .

[zhouyuan]

@malinjawi also run into this issue, it's due to the macos based timezone, the quick fix would be change the default value for timezone in gluten config. Please help to raise a fix.

[rui-mo]

@malinjawi @zhouyuan This is likely due to a bug in Folly (see facebookincubator/velox#16507) on MacOS. As a workaround, I’ve updated utcSet.find(zone) != utcSet.end(); to utcSet.find(std::string(zone)) != utcSet.end(); on my dev setup.
There is also another core dump issue in StringIdMap::release, which can be addressed as described here: facebookincubator/velox#16507 (comment).

[jinchengchenghh]

This issue only occurs in MAC environment, you could update the folly map and set to std::xxx standard API in your local.
But I'm not sure if Velox could accept it since the performance especially in cache is a key consideration.

[jinchengchenghh]

I created one facebookincubator/velox#17010, let wait for Velox community feedback