cshuo opened a new pull request, #18953:
URL: https://github.com/apache/hudi/pull/18953

   …flink1.20
   
   ### Describe the issue this Pull Request addresses
   
   The `hudi-flink1.20.x` module's `ParquetSplitReaderUtil` and its nested-type 
vector readers (`ArrayColumnReader`, `MapColumnReader`, `RowColumnReader`, 
`ArrayGroupReader`, and the `HeapArrayGroupColumnVector`) used Hudi's older, 
ad-hoc nested-Parquet decoding logic. Flink 2.1  line has since adopted a more 
robust Dremel-style striping/assembly reader (FLINK-35702) for 
`ARRAY`/`MAP`/`MULTISET`/`ROW` columns, which correctly handles 
repetition/definition levels, deeply nested structures, and missing/evolved 
fields. This PR backports that Dremel-based nested reader implementation to 
`hudi-flink1.20.x` so the module benefits from the same correctness 
improvements and stays aligned with the newer Flink versions Hudi already 
supports.
   
   ### Summary and Changelog
   
   The entire change is a verbatim copy of the corresponding files from 
hudi-flink1.19.x into hudi-flink1.20.x.
   
   - Vendors the Dremel-style nested Parquet reader from Apache Flink 2.1 
(FLINK-35702) into `hudi-flink1.20.x`, adding `NestedColumnReader`, 
`NestedPrimitiveColumnReader`, and the 
`ParquetField`/`ParquetGroupField`/`ParquetPrimitiveField` type model under 
`vector/type`.
   - Adds supporting position-tracking utilities: `CollectionPosition`, 
`LevelDelegation`, `RowPosition` (under `vector/position`), and 
`BooleanArrayList`, `IntArrayList`, `LongArrayList`, `NestedPositionUtil` 
(under `utils`) for tracking repetition/definition levels during 
striping/assembly.
   - Removes the legacy nested-type readers `ArrayColumnReader`, 
`ArrayGroupReader`, `MapColumnReader`, `RowColumnReader`, and the 
`HeapArrayGroupColumnVector`/`ColumnarGroupArrayData`/`ColumnarGroupMapData`/`ColumnarGroupRowData`
 vector classes, replacing them with the new Dremel-based reader and 
`HeapArrayVector`/`HeapMapColumnVector`/`HeapRowColumnVector`.
   - Reworks `ParquetSplitReaderUtil` to build a `ParquetField` tree per split 
via a new `buildFieldsList(List, List, MessageColumnIO)` helper and to delegate 
`ARRAY`/`MAP`/`MULTISET`/`ROW` columns to `NestedColumnReader`, while keeping 
Hudi's specialized primitive readers (`Int64TimestampColumnReader`, 
`FixedLenBytesColumnReader`, `HeapDecimalVector`) for top-level primitives 
unchanged.
   - Generalizes `createVectorFromConstant` (now `public`) to build 
constant/null-filled vectors for `ARRAY`, `MAP`, `MULTISET`, and `ROW` types 
(including recursively-constructed `HeapRowColumnVector` children), used both 
for partition-column injection and for filling missing nested fields with nulls 
so the Dremel assembler can pass through those slots.
   - Extends `HeapMapColumnVector`, `HeapRowColumnVector`, and 
`ParquetDecimalVector` with new constructors/accessors needed by the new reader 
path, and adds `MULTISET` and `TIMESTAMP_WITH_LOCAL_TIME_ZONE` handling 
alongside existing logical types.
   - Adds an overloaded `createColumnReader(..., ParquetField)` entry point and 
marks the previous primitive-only `createColumnReader` overload as 
`@Deprecated`, forwarding to the new field-aware path.
   - Adds new unit tests: `TestHeapColumnVectorAccessors`, 
`TestParquetDecimalVector`, `TestParquetDataColumnReaderFactory`, and 
`TestParquetGroupField`.
   
   ### Impact
   
   Reading `ARRAY`, `MAP`, `MULTISET`, and `ROW` (including nested/repeated) 
columns from Parquet via `hudi-flink1.20.x` now goes through the Dremel-style 
striping/assembly reader instead of the previous custom nested readers. 
Top-level primitive column reading is unchanged. No changes to public 
table/connector APIs.
   
   ### Risk Level
   
   medium - this replaces the core nested-type (ARRAY/MAP/MULTISET/ROW) Parquet 
column reading path in `hudi-flink1.20.x`, touching read correctness for any 
table with nested columns on Flink 1.20. New unit tests 
(`TestHeapColumnVectorAccessors`, `TestParquetDecimalVector`, 
`TestParquetDataColumnReaderFactory`, `TestParquetGroupField`) cover the new 
vector accessors, decimal handling, reader factory, and field-tree 
construction; broader end-to-end validation with nested-column Flink tables on 
Flink 1.20 is recommended before merge.
   
   ### Documentation Update
   
   none
   
   ### Contributor's checklist
   
   - [ ] Read through [contributor's 
guide](https://hudi.apache.org/contribute/how-to-contribute)
   - [ ] Enough context is provided in the sections above
   - [ ] Adequate tests were added if applicable


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to