Add an example to inspect parquet files and dump row group and page level metadata information #20117
Conversation
github-actions
bot
added
libcudf
mhaseeb123
changed the title
| * @brief Compute page row counts and page row offsets and column chunk page (count) offsets for a | ||
| * given column index | ||
| */ | ||
| [[nodiscard]] auto compute_page_row_counts_and_offsets( |
There was a problem hiding this comment.
Function modified from page_index_filter.cu
|
|
||
| cudf::host_span<uint8_t const> fetch_footer_bytes(cudf::host_span<uint8_t const> buffer) | ||
| { | ||
| CUDF_FUNC_RANGE(); |
There was a problem hiding this comment.
These fetch_xx functions copied as-is from hybrid_scan_common.cpp
| page_index_bytes.size()); | ||
| } | ||
|
|
||
| std::tuple<cudf::io::parquet::FileMetaData, bool> read_parquet_metadata( |
There was a problem hiding this comment.
Function modified from hybrid_scan_common.cpp
| return std::tuple{reader->parquet_metadata(), has_page_index}; | ||
| } | ||
|
|
||
| void write_rowgroup_metadata(cudf::io::parquet::FileMetaData const& metadata, |
There was a problem hiding this comment.
Function modified from hybrid_scan_helpers.cpp and reader_impl_helpers.cpp
mhaseeb123
added
feature request
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
mhaseeb123
added
3 - Ready for Review
| @@ -0,0 +1,422 @@ | |||
| /* | |||
There was a problem hiding this comment.
Most of the code in this file has been modified from libcudf internals
| @@ -0,0 +1,42 @@ | |||
| # Copyright (c) 2024-2025, NVIDIA CORPORATION. | |||
There was a problem hiding this comment.
Modified from the CMakeLists.txt in parquet_io example
mhaseeb123
added
5 - Ready to Merge
|
/merge |
Description
This PR adds a new libcudf example called
parquet_inspectwhich extracts useful row group and page level metadata information from the input parquet file's footer and page index (if available) and writes it to respective new parquet files.At the row group level, the written file contains three columns containing starting row offset, row counts and byte offset within the file for each row group
At the page-level, the written file contains three columns per input column (column in the input file) containing lists of page-level row offsets, lists of page-level row counts, and lists of page-level byte offsets within the file, one list per row group.
Note that the page-level metadata is only extracted and written if the page index is available in the parquet file.
Checklist