NVIDIA / NVIDIA/cudf

[FEA] Consider optionally rescaling literals when building Parquet predicate pushdown trees

Open
#22,217 0 comments 0 reactions 0 assignees View on GitHub
cuIO feature request libcudf
Dominant language
C++
Stars
9.8k
Forks
1.1k
Avg merge
3d 6m
Merged PRs (30d)
278

Description

## Description

In the parquet reader, we currently rescale column chunk and page stats to the target timestamp precision (if mismatched) during Predicate pushdown for correct comparisons. This also results in disabling bloom filter based pushdown for such columns.

Instead, we can consider constructing and pushing rescaled input literals to the converted expression trees which would undo changes in #22166 and also allow bloom filtering for mismatched precision timestamp columns.

That said, it's not very common to use bloom filter based row group pruning for numeric columns.

## Alternatives

Changes in PR #22166

## More context
_Originally posted by @mhaseeb123 in https://github.com/rapidsai/cudf/issues/22166#issuecomment-4283345379_

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.