Skip to content
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

Jump to bottom

Improve performance reading ByteViewArray from parquet by removing an implicit copy #6031

Merged
merged 2 commits into from
Jul 10, 2024
Merged

Improve performance reading ByteViewArray from parquet by removing an implicit copy #6031

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f250369
update byte view array to not implicit copy
XiangpengHao Jul 9, 2024
53d2fd6
Add small comments
XiangpengHao Jul 9, 2024
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
11 changes: 8 additions & 3 deletions parquet/src/arrow/array_reader/byte_view_array.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,6 @@ struct ByteViewArrayReader {
}

impl ByteViewArrayReader {
#[allow(unused)]
Copy link
Contributor

Choose a reason for hiding this comment

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

👍

fn new(
pages: Box<dyn PageIterator>,
data_type: ArrowType,
Expand Down Expand Up @@ -316,7 +315,10 @@ impl ByteViewArrayDecoderPlain {
}

pub fn read(&mut self, output: &mut ViewBuffer, len: usize) -> Result<usize> {
let block_id = output.append_block(self.buf.clone().into());
// Here we convert `bytes::Bytes` into `arrow_buffer::Bytes`, which is zero copy
// Then we convert `arrow_buffer::Bytes` into `arrow_buffer:Buffer`, which is also zero copy
let buf = arrow_buffer::Buffer::from_bytes(self.buf.clone().into());
Copy link
Contributor

Choose a reason for hiding this comment

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

I think we should at least add a comment the rationale for this non obvious code.

Maybe it would make sense to pull it into its a function (that could be commented, and more easily discoverable)

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Thinking about creating a from_arrow_bytes(...) and from_bytes(...), and then remove the impl<T: AsRef<[u8]>> From<T> for Buffer as it is too easy to misuse.

Copy link
Contributor

Choose a reason for hiding this comment

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

I recommend we split the code into multiple PRs -- this one to improve performance of the parquet reader and one to make it harder to misuse the API (which I suspect will be a breaking API change)

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Indeed, the breakage is much larger than I thought, will continue on this tomorrow.

Copy link
Contributor

Choose a reason for hiding this comment

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

Filed #6033 to track the idea

let block_id = output.append_block(buf);

let to_read = len.min(self.max_remaining_values);

Expand Down Expand Up @@ -546,7 +548,10 @@ impl ByteViewArrayDecoderDeltaLength {

let src_lengths = &self.lengths[self.length_offset..self.length_offset + to_read];

let block_id = output.append_block(self.data.clone().into());
// Here we convert `bytes::Bytes` into `arrow_buffer::Bytes`, which is zero copy
// Then we convert `arrow_buffer::Bytes` into `arrow_buffer:Buffer`, which is also zero copy
let bytes = arrow_buffer::Buffer::from_bytes(self.data.clone().into());
let block_id = output.append_block(bytes);

let mut current_offset = self.data_offset;
let initial_offset = current_offset;
Expand Down
1 change: 0 additions & 1 deletion parquet/src/arrow/buffer/view_buffer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,7 +68,6 @@ impl ViewBuffer {
}

/// Converts this into an [`ArrayRef`] with the provided `data_type` and `null_buffer`
#[allow(unused)]
pub fn into_array(self, null_buffer: Option<Buffer>, data_type: &ArrowType) -> ArrayRef {
let len = self.views.len();
let views = Buffer::from_vec(self.views);
Expand Down
Loading