Fix parquet complex type handling

[jaystarshot]

Fixes #7776

Parquet has notion of optional and repeated layers which is needed in arrow calls like DefLevelsToBitmap.

This info is passed using arrow:LevelInfo. We were incorrectly computing repeatedAncestor by ignoring optional fields which is fixed in this PR.

Parquet has 3 level structure for nested types
https://github.com/apache/parquet-format/blob/master/LogicalTypes.md#lists

// List<String> (list non-null, elements nullable)
1. required group my_list (LIST) {
2.   repeated group list {
3.    optional binary element (UTF8);
  }
}

However when we read this and convert to ParquetTypeWithId in current velox parquet reader, we ignore the intermediated layer 2. repeated group list (grandfather logic) in https://github.com/facebookincubator/velox/pull/9187/files#diff-64787e76c1b0ad12b5764770a94acd62054896a762ccead8f083a71a060f2f44R325.

[facebook-github-bot]

Hi @jaystarshot!

Thank you for your pull request.

We require contributors to sign our Contributor License Agreement, and yours needs attention.

You currently have a record in our system, but the CLA is no longer valid, and will need to be resubmitted.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[netlify]

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	231caed
🔍 Latest deploy log	https://app.netlify.com/sites/meta-velox/deploys/662c0be7c27b070008f773dc

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[jaystarshot]

Roughly based on https://github.com/apache/arrow/blob/main/cpp/src/parquet/level_conversion.h#L125.
the do() is to make sure that for nested types we are going inside and checking optional since the child layers optional (which we ignored) is needed by arrow

[jaystarshot]

cc: @Yuhta @yingsu00 @pranjalssh @rui-mo

[qqibrow]

could you elaborate more? it's a bit confusing that quote "This level will not be repeated" and next line isRepeated = true

[hitarth]

Remove the comment about result being incorrect now that it is fixed.

[qqibrow]

Please ensure all related DISABLED_ testes are reenabled. Also, in case we want a UT in ParquetReaderTest.cpp as well is a unit test here https://github.com/facebookincubator/velox/pull/8425/files

[jaystarshot]

It does pass locally for me with asan enabled.
I added

set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=address")
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=address")

to the makefile, and needed to add libasan.so.8 to the ldconfig

[Yuhta]

@jaystarshot Try debug build, it seems the OSS one is only running parquet tests with only release build

[jaystarshot]

Couldn't repro locally with the debug build as well
ctest -j ${NUM_THREADS} -VV --output-on-failure -R "parquet_table_scan" command used to run

[Yuhta]

Maybe add a VELOX_CHECK_LT(type.column(), fileMetaData_->row_groups[rowGroupIndex].columns.size()) before the crash point:

int64_t ReaderBase::rowGroupUncompressedSize(
    int32_t rowGroupIndex,
    const dwio::common::TypeWithId& type) const {
  if (type.column() != ParquetTypeWithId::kNonLeaf) {
    VELOX_CHECK_LT(type.column(), fileMetaData_->row_groups[rowGroupIndex].columns.size());
    return fileMetaData_->row_groups[rowGroupIndex]
        .columns[type.column()]
        .meta_data.total_uncompressed_size;
  }
  int64_t sum = 0;
  for (auto child : type.getChildren()) {
    sum += rowGroupUncompressedSize(rowGroupIndex, *child);
  }
  return sum;
}

Apparently the type has more columns than the metadata has

[jaystarshot]

Was the array test the only failure? If so it could be that there could be some issue in the test file and I can remove this test.

[Yuhta]

@jaystarshot It's both array and reqArrayLegacy

[jaystarshot]

I am wondering if this could be related to the crash #9223 is fixing.

[jaystarshot]

We can either keep those tests disabled in this PR or cherry pick that commit in this same PR and try to merge both

[jaystarshot]

@Yuhta Can you please try with this?

[qqibrow]

@jaystarshot I was about to add unit test for #9223 . Let me know whether I should proceed.

[jaystarshot]

Lets check if this fixes the crashes observed in meta's environment, if they do then we may not need tests

[Yuhta]

Can you rebase to latest to fix the build?

[facebook-github-bot]

@Yuhta has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[facebook-github-bot]

@Yuhta has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[facebook-github-bot]

@Yuhta merged this pull request in 37f4700.

[conbench-facebook]

Conbench analyzed the 1 benchmark run on commit 37f4700b.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details.