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.

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

Minor: Improve parquet PageIndex documentation #6042

Merged
merged 5 commits into from
Jul 17, 2024
Merged
Show file tree
Hide file tree
Changes from 3 commits
Commits
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
36 changes: 31 additions & 5 deletions parquet/src/file/metadata/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -50,17 +50,27 @@ use crate::schema::types::{
Type as SchemaType,
};

/// [`Index`] for each row group of each column.
/// Page level statistics for each column chunk of each row group.
///
/// This structure is an memory representation of multiple [`ColumnIndex`]
/// structures in a parquet file footer, as described in the Parquet [PageIndex
/// documentation]. Each [`Index`] holds statistics about all the pages in a
/// particular column chunk.
alamb marked this conversation as resolved.
Show resolved Hide resolved
///
/// `column_index[row_group_number][column_number]` holds the
/// [`Index`] corresponding to column `column_number` of row group
/// `row_group_number`.
///
/// For example `column_index[2][3]` holds the [`Index`] for the forth
/// column in the third row group of the parquet file.
///
/// [PageIndex documentation]: https://github.com/apache/parquet-format/blob/master/PageIndex.md
pub type ParquetColumnIndex = Vec<Vec<Index>>;

/// [`PageLocation`] for each data page of each row group of each column.
/// [`PageLocation`] for each data page of each row group of each column
///
/// This structure is the parsed representation of the [`OffsetIndex`] from the
/// Parquet file footer, as described in the Parquet [PageIndex documentation].
///
/// `offset_index[row_group_number][column_number][page_number]` holds
/// the [`PageLocation`] corresponding to page `page_number` of column
Expand All @@ -69,6 +79,8 @@ pub type ParquetColumnIndex = Vec<Vec<Index>>;
/// For example `offset_index[2][3][4]` holds the [`PageLocation`] for
/// the fifth page of the forth column in the third row group of the
/// parquet file.
///
/// [PageIndex documentation]: https://github.com/apache/parquet-format/blob/master/PageIndex.md
pub type ParquetOffsetIndex = Vec<Vec<Vec<PageLocation>>>;

/// Parsed metadata for a single Parquet file
Expand Down Expand Up @@ -942,14 +954,21 @@ impl ColumnChunkMetaDataBuilder {
}
}

/// Builder for column index
/// Builder for Parquet [`ColumnIndex`], part of the Parquet [PageIndex]
///
/// [PageIndex]: https://github.com/apache/parquet-format/blob/master/PageIndex.md
pub struct ColumnIndexBuilder {
null_pages: Vec<bool>,
min_values: Vec<Vec<u8>>,
max_values: Vec<Vec<u8>>,
null_counts: Vec<i64>,
boundary_order: BoundaryOrder,
// If one page can't get build index, need to ignore all index in this column
/// Is the information in the builder valid?
///
/// Set to `false` if any entry in the page doesn't have statistics for
/// some reason. This might happen if the page is entirely null, or
/// is a floating point column without any non-nan values
/// e.g. <https://github.com/apache/parquet-format/pull/196>
alamb marked this conversation as resolved.
Show resolved Hide resolved
valid: bool,
}

Expand All @@ -971,6 +990,7 @@ impl ColumnIndexBuilder {
}
}

/// Append statistics for the next page
pub fn append(
&mut self,
null_page: bool,
Expand All @@ -988,15 +1008,19 @@ impl ColumnIndexBuilder {
self.boundary_order = boundary_order;
}

/// Mark this column index as invalid
pub fn to_invalid(&mut self) {
self.valid = false;
}

/// Is the information in the builder valid?
pub fn valid(&self) -> bool {
self.valid
}

/// Build and get the thrift metadata of column index
///
/// Note: callers should check [`Self::valid`] before calling this method
Copy link
Contributor

Choose a reason for hiding this comment

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

Makes me wonder if build_to_thrift should return an Option, to make sure callers don't forget to check .valid().

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I agree that would be a good improvement -- filed #6064

pub fn build_to_thrift(self) -> ColumnIndex {
ColumnIndex::new(
self.null_pages,
Expand All @@ -1008,7 +1032,9 @@ impl ColumnIndexBuilder {
}
}

/// Builder for offset index
/// Builder for offset index, part of the Parquet [PageIndex].
///
/// [PageIndex]: https://github.com/apache/parquet-format/blob/master/PageIndex.md
pub struct OffsetIndexBuilder {
offset_array: Vec<i64>,
compressed_page_size_array: Vec<i32>,
Expand Down
34 changes: 21 additions & 13 deletions parquet/src/file/page_index/index.rs
Original file line number Diff line number Diff line change
Expand Up @@ -25,14 +25,9 @@ use crate::format::{BoundaryOrder, ColumnIndex};
use crate::util::bit_util::from_le_slice;
use std::fmt::Debug;

/// PageIndex Statistics for one data page, as described in [Column Index].
/// Typed statistics for one data page
///
/// One significant difference from the row group level
/// [`Statistics`](crate::format::Statistics) is that page level
/// statistics may not store actual column values as min and max
/// (e.g. they may store truncated strings to save space)
///
/// [Column Index]: https://github.com/apache/parquet-format/blob/master/PageIndex.md
/// See [`NativeIndex`] for more details
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct PageIndex<T> {
/// The minimum value, It is None when all values are null
Expand Down Expand Up @@ -70,11 +65,9 @@ where

#[derive(Debug, Clone, PartialEq)]
#[allow(non_camel_case_types)]
/// Typed statistics for a data page in a column chunk.
/// Statistics for data pages in a column chunk.
///
/// This structure is part of the "Page Index" and is optionally part of
/// [ColumnIndex] in the parquet file and can be used to skip decoding pages
/// while reading the file data.
/// See [NativeIndex] for more information
alamb marked this conversation as resolved.
Show resolved Hide resolved
pub enum Index {
/// Sometimes reading page index from parquet file
/// will only return pageLocations without min_max index,
Expand Down Expand Up @@ -117,10 +110,25 @@ impl Index {
}
}

/// Stores the [`PageIndex`] for each page of a column
/// Strongly typed statistics for data pages in a column chunk.
///
/// This structure is a natively typed, in memory representation of the
/// [`ColumnIndex`] structure in a parquet file footer, as described in the
/// Parquet [PageIndex documentation]. The statistics stored in this structure
/// can be used by query engines to skip decoding pages while reading parquet
/// data.
///
/// # Differences with Row Group Level Statistics
///
/// One significant difference between `NativeIndex` and row group level
/// [`Statistics`] is that page level statistics may not store actual column
/// values as min and max (e.g. they may store truncated strings to save space)
///
/// [PageIndex documentation]: https://github.com/apache/parquet-format/blob/master/PageIndex.md
/// [`Statistics`]: crate::file::statistics::Statistics
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct NativeIndex<T: ParquetValueType> {
/// The indexes, one item per page
/// The actual column indexes, one item per page
pub indexes: Vec<PageIndex<T>>,
/// If the min/max elements are ordered, and if so in which
/// direction. See [source] for details.
Expand Down
16 changes: 14 additions & 2 deletions parquet/src/file/statistics.rs
Original file line number Diff line number Diff line change
Expand Up @@ -287,7 +287,17 @@ pub fn to_thrift(stats: Option<&Statistics>) -> Option<TStatistics> {
Some(thrift_stats)
}

/// Statistics for a column chunk and data page.
/// Strongly typed statistics for a column chunk within a row group.
///
/// This structure is a natively typed, in memory representation of the
/// [`Statistics`] structure in a parquet file footer. The statistics stored in
/// this structure can be used by query engines to skip decoding pages while
/// reading parquet data.
///
/// Page level statistics are stored separately, in [NativeIndex].
///
/// [`Statistics`]: crate::format::Statistics
/// [NativeIndex]: crate::file::page_index::index::NativeIndex
#[derive(Debug, Clone, PartialEq)]
pub enum Statistics {
Boolean(ValueStatistics<bool>),
Expand Down Expand Up @@ -445,7 +455,9 @@ impl fmt::Display for Statistics {
/// Typed implementation for [`Statistics`].
pub type TypedStatistics<T> = ValueStatistics<<T as DataType>::T>;

/// Statistics for a particular `ParquetValueType`
/// Typed statistics for one column chunk
///
/// See [`Statistics`] for more details
#[derive(Clone, Eq, PartialEq)]
pub struct ValueStatistics<T> {
min: Option<T>,
Expand Down
Loading