danny0405 opened a new pull request, #19293: URL: https://github.com/apache/hudi/pull/19293
### Describe the issue this Pull Request addresses `FileSlice.getTotalFileSizeAsParquetFormat` previously applied the configured log-to-Parquet compression ratio to every log file in a file slice. This incorrectly recalibrated native log files and inline log files containing non-Avro data blocks, even though their physical sizes already represent the expected size. `DeltaWriteProfile` also applied the compression ratio without accounting for the table version. Table version 10 uses native logs, so calibration should only apply to tables below version 10 when the effective inline data block type is Avro. ### Summary and Changelog This change limits log-to-Parquet size calibration to legacy inline Avro log files and preserves the physical size for native and non-Avro log files. #### Working tree: calibrate only inline Avro log files - Changed `FileSlice.getTotalFileSizeAsParquetFormat` to accept `HoodieConfig` instead of a compression-ratio scalar. - Separately sums inline and native log file sizes. - Applies the compression ratio once to the total inline log size when the configured data block format is Avro. - Uses physical sizes for native log files and inline Parquet/HFile data blocks. - Updated Spark and Flink callers, including `SparkUpsertDeltaCommitPartitioner`, `DeltaWriteProfile`, `HoodieFileIndex`, and `PartitionDirectoryConverter`. - Updated `DeltaWriteProfile.logFileToParquetCompressionRatio` to return the configured ratio only when the write table version is below 10 and the effective log data block type is Avro; all other cases return `1`. - Added `TestFileSlice` coverage for mixed inline/native logs and Avro, Parquet, and HFile block formats. - Added `TestBucketAssigner` coverage for version 9 inline logs and version 10 native logs. - Updated `TestPartitionDirectoryConverter` for the configuration-based API. ### Impact This improves small-file detection, bucket sizing, average-record-size estimation, and Spark file-split estimation for Merge-On-Read tables using native or non-Avro log files. There are no storage-format, configuration-default, or runtime compatibility changes. The public `FileSlice.getTotalFileSizeAsParquetFormat` method signature changes from accepting a compression ratio to accepting `HoodieConfig`; all in-repository callers are updated, but direct downstream callers may require a source update. ### Risk Level medium The change affects shared file-slice size estimation used by Spark and Flink write/read paths. Risk is mitigated by targeted coverage and successful validation: - `TestFileSlice`: 7 tests passed. - `TestBucketAssigner`: 17 tests passed, including inline Avro, inline Parquet, and native-log scenarios. - The affected Flink dependency reactor completed successfully. - The affected Spark, Flink, and common modules compiled successfully. - `git diff --check` passed. ### Documentation Update none This corrects internal size-estimation behavior without introducing a new configuration, changing a default, or adding a user-facing feature. ### 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]
