[Parquet] Support skipping pages with mask based evaluation #9118
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
sdf-jkl
changed the title
Contributor
Author
alamb
reviewed
Jan 10, 2026
Contributor
alamb
left a comment
alamb
left a comment
There was a problem hiding this comment.
Thank you @sdf-jkl -- this actually makes a lot of sense to me 👏
I have a few concerns:
- I am worried about the performance overhead of this approach (copying the page index and the loop for each batch) -- I will run some benchmarks to asses this
- I do wonder if we have test coverage for this entire situation -- in particular, do we have tests that repeatedly call
next_mask_chunkafter the first page and make sure we get the right rows?
If the performance looks good, I think we should add some more tests -- maybe @hhhizzz has some ideas on how to do this (or I think I can try and find some time to help out / work with codex to do so)
| /// Using the row selection to skip(4), page2 won't be read at all, so in this | ||
| /// case we can't decode all the rows and apply a mask. To correctly apply the | ||
| /// bit mask, we need all 6 values be read, but page2 is not in memory. | ||
| fn override_selector_strategy_if_needed( |
Contributor
There was a problem hiding this comment.
nice -- the idea is to avoid this function 👍
| array_reader, | ||
| schema: Arc::new(schema), | ||
| read_plan, | ||
| page_offsets: page_offsets.map(|slice| Arc::new(slice.to_vec())), |
Contributor
There was a problem hiding this comment.
So I think this will effectively will copy the entire OffsetIndexMetadataData structure (which I worry could be quite large)
I wonder if we need to find a way to avoid this (e.g. making the entire thing Arc'd in https://github.com/apache/arrow-rs/blob/67e04e758f1e62ec3d78d2f678daf433a4c54e30/parquet/src/file/metadata/mod.rs#L197-L196 somehow 🤔 )
Contributor
Author
There was a problem hiding this comment.
We could store only the &Vec<PageLocation> instead of the entire OffsetIndexMetadataData df9a493
| while cursor < mask.len() && selected_rows < batch_size { | ||
| let mut page_end = mask.len(); | ||
| if let Some(pages) = page_locations { | ||
| for loc in pages { |
Contributor
There was a problem hiding this comment.
I am also a little worried that this loop will take too long (it is O(N^2) in the number of pages as each time it looks through all pages
Maybe we could potentially add a PageLocationIterator to the cursor itself (so we know where to pick up)
Contributor
Author
There was a problem hiding this comment.
Maybe a binary search through a vec of page offsets? Would have to construct the vec once beforehand to keep us from rebuilding it.
Contributor
|
run benchmark arrow_reader_clickbench arrow_reader_row_filter |
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
Contributor
|
run benchmark arrow_reader_clickbench arrow_reader_row_filter |
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Which issue does this PR close?
Rationale for this change
Check issue.
What changes are included in this PR?
Made
next_mask_chunkpage aware.By adding
page_offsetstoParquetRecordBatchReaderAre these changes tested?
Should be covered by existing tests from #8733
Are there any user-facing changes?