Skip to content

Parquet: Do not compress v2 data page when compress is bad quality #8257

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
alamb merged 3 commits into from
Sep 8, 2025
Merged

Parquet: Do not compress v2 data page when compress is bad quality #8257

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e0077f9
Do not compress v2 data page when compress is bad quality
mapleFU Sep 1, 2025
2d66332
Merge branch 'main' into page-v2-no-compress-if-size-too-lage
mapleFU Sep 5, 2025
0e198e9
Add testing
mapleFU Sep 5, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 44 additions & 4 deletions parquet/src/column/writer/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1104,12 +1104,23 @@ impl<'a, E: ColumnValueEncoder> GenericColumnWriter<'a, E> {
rep_levels_byte_len + def_levels_byte_len + values_data.buf.len();

// Data Page v2 compresses values only.
match self.compressor {
let is_compressed = match self.compressor {
Some(ref mut cmpr) => {
let buffer_len = buffer.len();
cmpr.compress(&values_data.buf, &mut buffer)?;
if uncompressed_size <= buffer.len() - buffer_len {
Copy link
Member Author

@mapleFU mapleFU Sep 1, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This can also be a "score", like if uncompressed_size <= (buffer.len() - buffer_len) * 0.9 {

Copy link
Contributor

@etseidl etseidl Sep 4, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd suggest maybe going the opposite direction. If compression doesn't give say a 10% size reduction, then don't compress...it will be faster to read. 10% might be too much...maybe we could make it configurable.

Anyway, that's probably out of scope for this PR. Let's just go with the simple test for now...it's still an improvement.

Copy link
Contributor

@alamb alamb Sep 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, I do think that would need to configurable as I think some people would treat a 10% size increase as a regression, even if they got a faster reader

A good follow on ticket for sure

Copy link
Contributor

@JigaoLuo JigaoLuo Sep 9, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I’ll follow up once there’s a plan for this future ticket. Just a note from me —the score is still a highly empirical value (and depends on the decompression speed), so I’ve embedded it into my rewriter and have been extensively rewriting Parquet files based on it.

Copy link
Contributor

@alamb alamb Sep 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In my opinion, given there is a tradeoff here (file size / decode speed) we can't hard code some heuristic that changes the default behavior.

However, I think it would be a good improvement to add some sort of tuning knob for people who wanted to make that tradeoff more explicitly as long as the default setting of the knob leaves the existing behavior

buffer.truncate(buffer_len);
buffer.extend_from_slice(&values_data.buf);
false
} else {
true
}
}
None => buffer.extend_from_slice(&values_data.buf),
}
None => {
buffer.extend_from_slice(&values_data.buf);
false
}
};

let data_page = Page::DataPageV2 {
buf: buffer.into(),
Expand All @@ -1119,7 +1130,7 @@ impl<'a, E: ColumnValueEncoder> GenericColumnWriter<'a, E> {
num_rows: self.page_metrics.num_buffered_rows,
def_levels_byte_len: def_levels_byte_len as u32,
rep_levels_byte_len: rep_levels_byte_len as u32,
is_compressed: self.compressor.is_some(),
is_compressed,
statistics: page_statistics,
};

Expand Down Expand Up @@ -4236,4 +4247,33 @@ mod tests {
.unwrap();
ColumnDescriptor::new(Arc::new(tpe), max_def_level, max_rep_level, path)
}

#[test]
fn test_page_v2_snappy_compression_fallback() {
// Test that PageV2 sets is_compressed to false when Snappy compression increases data size
let page_writer = TestPageWriter {};

// Create WriterProperties with PageV2 and Snappy compression
let props = WriterProperties::builder()
.set_writer_version(WriterVersion::PARQUET_2_0)
// Disable dictionary to ensure data is written directly
.set_dictionary_enabled(false)
.set_compression(Compression::SNAPPY)
.build();

let mut column_writer =
get_test_column_writer::<ByteArrayType>(Box::new(page_writer), 0, 0, Arc::new(props));

// Create small, simple data that Snappy compression will likely increase in size
// due to compression overhead for very small data
Copy link
Contributor

@alamb alamb Sep 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

👍

let values = vec![ByteArray::from("a")];

column_writer.write_batch(&values, None, None).unwrap();

let result = column_writer.close().unwrap();
assert_eq!(
result.metadata.uncompressed_size(),
result.metadata.compressed_size()
);
}
}
Loading