chenghuichen commented on PR #7673: URL: https://github.com/apache/paimon/pull/7673#issuecomment-4457753276
@leaves12138 Thanks for catching this! The `_ROW_ID` correctness issue makes total sense. Fixed both cases: - SlicedSplit: offset `first_row_id` by `shard_range[0]` for correct `_ROW_ID` - IndexedSplit: pass local offsets to `DataFileBatchReader` for per-row `_ROW_ID` assignment Added e2e regression tests (with_shard + `_ROW_ID` projection, IndexedSplit + `_ROW_ID`). Benchmark Result: (20M rows, 4 shards, interleaved A/B in same process): | Format | Full read + slice | Native pushdown | Speedup | |--------|-------------------|-----------------|---------| | **Lance** | 1719.7 ms | 735.3 ms | **57%** | | **Vortex** | 1905.2 ms | 638.4 ms | **67%** | Index materialization cost is negligible relative to the I/O savings. Script below for reproduction. [bench_shard_pushdown.py](https://github.com/user-attachments/files/27790887/bench_shard_pushdown.py) -- 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]
