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

Add parquet page stats for float{16, 32, 64} #10982

Merged
merged 1 commit into from
Jun 19, 2024
Merged

Add parquet page stats for float{16, 32, 64} #10982

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
100 changes: 92 additions & 8 deletions datafusion/core/src/datasource/physical_plan/parquet/statistics.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,8 +33,9 @@ use arrow_array::{
use arrow_schema::{Field, FieldRef, Schema, TimeUnit};
use datafusion_common::{internal_datafusion_err, internal_err, plan_err, Result};
use half::f16;
use parquet::data_type::FixedLenByteArray;
use parquet::file::metadata::{ParquetColumnIndex, ParquetOffsetIndex, RowGroupMetaData};
use parquet::file::page_index::index::Index;
use parquet::file::page_index::index::{Index, PageIndex};
use parquet::file::statistics::Statistics as ParquetStatistics;
use parquet::schema::types::SchemaDescriptor;
use paste::paste;
Expand Down Expand Up @@ -495,7 +496,7 @@ macro_rules! get_statistics {
}

macro_rules! make_data_page_stats_iterator {
($iterator_type: ident, $func: ident, $index_type: path, $stat_value_type: ty) => {
($iterator_type: ident, $func: expr, $index_type: path, $stat_value_type: ty) => {
struct $iterator_type<'a, I>
where
I: Iterator<Item = (usize, &'a Index)>,
Expand Down Expand Up @@ -526,7 +527,7 @@ macro_rules! make_data_page_stats_iterator {
native_index
.indexes
.iter()
.map(|x| x.$func)
.map(|x| $func(x))
.collect::<Vec<_>>(),
),
// No matching `Index` found;
Expand All @@ -548,11 +549,66 @@ macro_rules! make_data_page_stats_iterator {
};
}

make_data_page_stats_iterator!(MinInt32DataPageStatsIterator, min, Index::INT32, i32);
make_data_page_stats_iterator!(MaxInt32DataPageStatsIterator, max, Index::INT32, i32);
make_data_page_stats_iterator!(MinInt64DataPageStatsIterator, min, Index::INT64, i64);
make_data_page_stats_iterator!(MaxInt64DataPageStatsIterator, max, Index::INT64, i64);

make_data_page_stats_iterator!(
MinInt32DataPageStatsIterator,
|x: &PageIndex<i32>| { x.min },
Index::INT32,
i32
);
make_data_page_stats_iterator!(
MaxInt32DataPageStatsIterator,
|x: &PageIndex<i32>| { x.max },
Index::INT32,
i32
);
make_data_page_stats_iterator!(
MinInt64DataPageStatsIterator,
|x: &PageIndex<i64>| { x.min },
Index::INT64,
i64
);
make_data_page_stats_iterator!(
MaxInt64DataPageStatsIterator,
|x: &PageIndex<i64>| { x.max },
Index::INT64,
i64
);
make_data_page_stats_iterator!(
MinFloat16DataPageStatsIterator,
|x: &PageIndex<FixedLenByteArray>| { x.min.clone() },
Index::FIXED_LEN_BYTE_ARRAY,
Copy link
Contributor

Choose a reason for hiding this comment

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

cargo doc --document-private-items -p datafusion --open

Screenshot 2024-06-18 at 4 32 10 PM

I see what is going on -- I think this makes sense for now

Implies that the iterator is an iterator over Vec<FixedLengthItem

Copy link
Contributor Author

Choose a reason for hiding this comment

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

exactly, basically caused by there being no f16 index in https://arrow.apache.org/rust/src/parquet/file/page_index/index.rs.html#74 ... I was thinking of mapping to f16 first earlier on, to make it cleaner... but this turned out to be the simplest I came up with.

#10972 (comment) is I think the best way to make this better eventually

FixedLenByteArray
);
make_data_page_stats_iterator!(
MaxFloat16DataPageStatsIterator,
|x: &PageIndex<FixedLenByteArray>| { x.max.clone() },
Index::FIXED_LEN_BYTE_ARRAY,
FixedLenByteArray
);
make_data_page_stats_iterator!(
MinFloat32DataPageStatsIterator,
|x: &PageIndex<f32>| { x.min },
Index::FLOAT,
f32
);
make_data_page_stats_iterator!(
MaxFloat32DataPageStatsIterator,
|x: &PageIndex<f32>| { x.max },
Index::FLOAT,
f32
);
make_data_page_stats_iterator!(
MinFloat64DataPageStatsIterator,
|x: &PageIndex<f64>| { x.min },
Index::DOUBLE,
f64
);
make_data_page_stats_iterator!(
MaxFloat64DataPageStatsIterator,
|x: &PageIndex<f64>| { x.max },
Index::DOUBLE,
f64
);
macro_rules! get_data_page_statistics {
($stat_type_prefix: ident, $data_type: ident, $iterator: ident) => {
paste! {
Expand Down Expand Up @@ -581,6 +637,19 @@ macro_rules! get_data_page_statistics {
)),
Some(DataType::Int32) => Ok(Arc::new(Int32Array::from_iter([<$stat_type_prefix Int32DataPageStatsIterator>]::new($iterator).flatten()))),
Some(DataType::Int64) => Ok(Arc::new(Int64Array::from_iter([<$stat_type_prefix Int64DataPageStatsIterator>]::new($iterator).flatten()))),
Some(DataType::Float16) => Ok(Arc::new(
Float16Array::from_iter(
[<$stat_type_prefix Float16DataPageStatsIterator>]::new($iterator)
.map(|x| {
x.into_iter().filter_map(|x| {
x.and_then(|x| Some(from_bytes_to_f16(x.data())))
})
})
.flatten()
)
)),
Some(DataType::Float32) => Ok(Arc::new(Float32Array::from_iter([<$stat_type_prefix Float32DataPageStatsIterator>]::new($iterator).flatten()))),
Some(DataType::Float64) => Ok(Arc::new(Float64Array::from_iter([<$stat_type_prefix Float64DataPageStatsIterator>]::new($iterator).flatten()))),
_ => unimplemented!()
}
}
Expand Down Expand Up @@ -677,6 +746,21 @@ where
.iter()
.map(|x| x.null_count.map(|x| x as u64))
.collect::<Vec<_>>(),
Index::FLOAT(native_index) => native_index
.indexes
.iter()
.map(|x| x.null_count.map(|x| x as u64))
.collect::<Vec<_>>(),
Index::DOUBLE(native_index) => native_index
.indexes
.iter()
.map(|x| x.null_count.map(|x| x as u64))
.collect::<Vec<_>>(),
Index::FIXED_LEN_BYTE_ARRAY(native_index) => native_index
.indexes
.iter()
.map(|x| x.null_count.map(|x| x as u64))
.collect::<Vec<_>>(),
_ => unimplemented!(),
});

Expand Down
88 changes: 88 additions & 0 deletions datafusion/core/tests/parquet/arrow_statistics.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -614,6 +614,94 @@ async fn test_int_8() {
.run();
}

#[tokio::test]
Copy link
Contributor

Choose a reason for hiding this comment

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

💯 very nice tests

Copy link
Contributor

@tshauck tshauck Jun 18, 2024 •
edited
Loading

Choose a reason for hiding this comment

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

I think this test is mostly a duplicate of test_float16 just that one has Check::RowGroup vs Check::Both... maybe one or the other can be updated?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

@tshauck this is one of the things that were bugging me a bit, the other one being https://github.com/apache/datafusion/pull/10982/files#diff-7110f4709c105a18ef74a212396444d62052179a735d148fb62470a8b157fb40R749-R763 -- both are very repetitive

however, I didn't want to get overeager, only to realize later than the abstractions chosen was not the right one. Perhaps the best way forward would be to address #10952 first (which may also have its own "float16"-like tricky case), and then getting the correct macros for testing, index handling, etc. Speaking of that, would you like to take that one? I'd be happy to review then

Copy link
Contributor

Choose a reason for hiding this comment

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

Sounds like a good plan to me -- let's merge this PR and then work on #10952

Copy link
Contributor

Choose a reason for hiding this comment

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

Filed #11000 to track duplication

async fn test_float_16() {
// This creates a parquet files of 1 column named f
let reader = TestReader {
scenario: Scenario::Float16,
row_per_group: 5,
}
.build()
.await;

Test {
reader: &reader,
// mins are [-5, -4, 0, 5]
expected_min: Arc::new(Float16Array::from(vec![
f16::from_f32(-5.),
f16::from_f32(-4.),
f16::from_f32(-0.),
f16::from_f32(5.),
])),
// maxes are [-1, 0, 4, 9]
expected_max: Arc::new(Float16Array::from(vec![
f16::from_f32(-1.),
f16::from_f32(0.),
f16::from_f32(4.),
f16::from_f32(9.),
])),
// nulls are [0, 0, 0, 0]
expected_null_counts: UInt64Array::from(vec![0, 0, 0, 0]),
// row counts are [5, 5, 5, 5]
expected_row_counts: Some(UInt64Array::from(vec![5, 5, 5, 5])),
column_name: "f",
check: Check::Both,
}
.run();
}

#[tokio::test]
async fn test_float_32() {
// This creates a parquet files of 1 column named f
let reader = TestReader {
scenario: Scenario::Float32,
row_per_group: 5,
}
.build()
.await;

Test {
reader: &reader,
// mins are [-5, -4, 0, 5]
expected_min: Arc::new(Float32Array::from(vec![-5., -4., -0., 5.0])),
// maxes are [-1, 0, 4, 9]
expected_max: Arc::new(Float32Array::from(vec![-1., 0., 4., 9.])),
// nulls are [0, 0, 0, 0]
expected_null_counts: UInt64Array::from(vec![0, 0, 0, 0]),
// row counts are [5, 5, 5, 5]
expected_row_counts: Some(UInt64Array::from(vec![5, 5, 5, 5])),
column_name: "f",
check: Check::Both,
}
.run();
}

#[tokio::test]
async fn test_float_64() {
Copy link
Contributor

Choose a reason for hiding this comment

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

There's a similar test_float64 in this file that could similarly be updated or removed?

// This creates a parquet files of 1 column named f
let reader = TestReader {
scenario: Scenario::Float64,
row_per_group: 5,
}
.build()
.await;

Test {
reader: &reader,
// mins are [-5, -4, 0, 5]
expected_min: Arc::new(Float64Array::from(vec![-5., -4., -0., 5.0])),
// maxes are [-1, 0, 4, 9]
expected_max: Arc::new(Float64Array::from(vec![-1., 0., 4., 9.])),
// nulls are [0, 0, 0, 0]
expected_null_counts: UInt64Array::from(vec![0, 0, 0, 0]),
// row counts are [5, 5, 5, 5]
expected_row_counts: Some(UInt64Array::from(vec![5, 5, 5, 5])),
column_name: "f",
check: Check::Both,
}
.run();
}

// timestamp
#[tokio::test]
async fn test_timestamp() {
Expand Down
15 changes: 15 additions & 0 deletions datafusion/core/tests/parquet/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,6 +90,7 @@ enum Scenario {
/// -MIN, -100, -1, 0, 1, 100, MAX
NumericLimits,
Float16,
Float32,
Float64,
Decimal,
Decimal256,
Expand Down Expand Up @@ -586,6 +587,12 @@ fn make_f64_batch(v: Vec<f64>) -> RecordBatch {
RecordBatch::try_new(schema, vec![array.clone()]).unwrap()
}

fn make_f32_batch(v: Vec<f32>) -> RecordBatch {
let schema = Arc::new(Schema::new(vec![Field::new("f", DataType::Float32, true)]));
let array = Arc::new(Float32Array::from(v)) as ArrayRef;
RecordBatch::try_new(schema, vec![array.clone()]).unwrap()
}

fn make_f16_batch(v: Vec<f16>) -> RecordBatch {
let schema = Arc::new(Schema::new(vec![Field::new("f", DataType::Float16, true)]));
let array = Arc::new(Float16Array::from(v)) as ArrayRef;
Expand Down Expand Up @@ -1003,6 +1010,14 @@ fn create_data_batch(scenario: Scenario) -> Vec<RecordBatch> {
),
]
}
Scenario::Float32 => {
vec![
make_f32_batch(vec![-5.0, -4.0, -3.0, -2.0, -1.0]),
make_f32_batch(vec![-4.0, -3.0, -2.0, -1.0, 0.0]),
make_f32_batch(vec![0.0, 1.0, 2.0, 3.0, 4.0]),
make_f32_batch(vec![5.0, 6.0, 7.0, 8.0, 9.0]),
]
}
Scenario::Float64 => {
vec![
make_f64_batch(vec![-5.0, -4.0, -3.0, -2.0, -1.0]),
Expand Down