[CI][Integration] Spark jobs are failing with problem on org.apache.arrow.flatbuf

[raulcd]

Describe the bug, including details regarding any error messages, version, and platform.

The spark jobs:
test-conda-python-3.10-spark-v3.5.0
test-conda-python-3.11-spark-master
test-conda-python-3.8-spark-v3.5.0

Are currently failing with:

 Error: ] /spark/sql/core/src/main/scala/org/apache/spark/sql/execution/arrow/ArrowConverters.scala:26: object flatbuf is not a member of package org.apache.arrow
Error: ] /spark/sql/core/src/main/scala/org/apache/spark/sql/execution/arrow/ArrowConverters.scala:456: Class org.apache.arrow.flatbuf.Message not found - continuing with a stub.
Error: [ERROR] two errors found
Error:  Failed to execute goal net.alchim31.maven:scala-maven-plugin:4.8.0:compile (scala-compile-first) on project spark-sql_2.12: Execution scala-compile-first of goal net.alchim31.maven:scala-maven-plugin:4.8.0:compile failed: org.apache.commons.exec.ExecuteException: Process exited with an error: 255 (Exit value: 255) -> [Help 1]
Error:  
Error:  To see the full stack trace of the errors, re-run Maven with the -e switch.
Error:  Re-run Maven using the -X switch to enable full debug logging.
Error:  
Error:  For more information about the errors and possible solutions, please read the following articles:
Error:  [Help 1] http://cwiki.apache.org/confluence/display/MAVEN/PluginExecutionException
Error:  
Error:  After correcting the problems, you can resume the build with the command
Error:    mvn <args> -rf :spark-sql_2.12

Component(s)

Continuous Integration, Integration

[raulcd]

This is also reproducible for 16.1.0 but I don't think this is related to any change on our side and doesn't seem to be a blocker.
@danepitkin @vibhatha @kiszk any thoughts?

[kiszk]

The error message is the same as #40549.

[kiszk]

2024-04-28, success
test-conda-python-3.10-spark-v3.5.0

2024-04-29, failure
test-conda-python-3.10-spark-v3.5.0

2024-05-07?, failure
test-conda-python-3.10-spark-v3.5.0

2024-05-09, failure
test-conda-python-3.10-spark-v3.5.0

2024-05-12, failure
test-conda-python-3.10-spark-v3.5.0

[kiszk]

In the previous issue, this PR causes the issue.

MINOR: [Java] Bump org.apache.maven.plugins:maven-shade-plugin from 3.2.4 to 3.5.2 in /java

This PR, which was merged on April 29, include the following change.

Update maven-shade-plugin to 3.5.1 (via org.apache:parent)

[raulcd]

@danepitkin @lidavidm what should we do with this one?

[kiszk]

#41309 (comment) causes this failure.

[kiszk]

I am just surprised this comment is completely ignored.

[lidavidm]

We can revert. Sorry about that @kiszk

[lidavidm]

It would be good to know why maven-shade-plugin is causing issues, though

[lidavidm]

Issue resolved by pull request 41628
#41628

[laurentgo]

I guess my bad but it's quite hard to work on improving the build and also track tests which are not part of the CI. I will repost the change though with hopefully a fix, although nobody really understands what's the exact issue?

[lidavidm]

Yes, I've asked Vibhatha if we can get more of those tests into proper CI...

[laurentgo]

The issue seems to be because maven-shade-plugin generates and attach a new pom file containing the following dependency:

    <dependency>
      <groupId>org.apache.arrow</groupId>
      <artifactId>arrow-format</artifactId>
      <version>17.0.0-SNAPSHOT</version>
      <scope>provided</scope>
    </dependency>

which is exactly what the plugin is supposed to be doing

I have the same resulting pom by using maven-shade-plugin 3.3.0, and going through changes, I found this change which is a bugfix to the plugin for this issue

As to how to address this within the Arrow project, I'd need some perspective. The main issue is that a Maven module has a unique pom.xml definition for all the artifacts attached to it. But when attaching shaded artifacts the list of dependencies is not quite the same, and this is definitively conflicting.
I have no idea which project is actually using the shaded version of arrow-vector, but Spark isn't and so if we agree that the pom.xml should reflect the dependencies for the main artifact, then we should tell maven-shade-plugin to not generate a new pom.xml file. If we need both dependencies, then we are quite in a predicament I'd guess (and same would apply to the other plugins using maven-shade-plugin)

[lidavidm]

Hmm, the POM should reflect the main file. The shaded version might need to be split into a separate artifact...