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

fix(cubesql): properly convert unsigned types from arrow to dataframe #9029

Open
wants to merge 5 commits into
base: master
Choose a base branch
from
Open

fix(cubesql): properly convert unsigned types from arrow to dataframe #9029

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
110e08e
fix(cubesql): properly convert type from arrow to dataframe
max-kammerer Dec 10, 2024
8b26796
rearrange
max-kammerer Dec 10, 2024
922651e
fix Decimal
max-kammerer Dec 10, 2024
0fa8776
~
max-kammerer Dec 10, 2024
f443d9b
fmt
max-kammerer Dec 10, 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
244 changes: 229 additions & 15 deletions rust/cubesql/cubesql/src/sql/dataframe.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -409,14 +409,15 @@ pub fn arrow_to_column_type(arrow_type: DataType) -> Result<ColumnType, CubeErro
DataType::Float16 | DataType::Float32 | DataType::Float64 => Ok(ColumnType::Double),
DataType::Boolean => Ok(ColumnType::Boolean),
DataType::List(field) => Ok(ColumnType::List(field)),
DataType::Int32 | DataType::UInt32 => Ok(ColumnType::Int32),
DataType::Decimal(_, _) => Ok(ColumnType::Int32),
DataType::Int8
| DataType::Int16
| DataType::Int64
| DataType::UInt8
DataType::Int8 //we are missing TableValue::Int8 type to use ColumnType:Int8
| DataType::UInt8 //we are missing ColumnType::Int16 type
| DataType::Int16 //we are missing ColumnType::Int16 type
| DataType::UInt16
| DataType::UInt64 => Ok(ColumnType::Int64),
| DataType::Int32 => Ok(ColumnType::Int32),
DataType::UInt32
| DataType::Int64 => Ok(ColumnType::Int64),
DataType::UInt64 => Ok(ColumnType::Decimal(39, 0)),
DataType::Null => Ok(ColumnType::String),
x => Err(CubeError::internal(format!("unsupported type {:?}", x))),
}
Expand Down Expand Up @@ -452,12 +453,23 @@ pub fn batches_to_dataframe(
let array = batch.column(column_index);
let num_rows = batch.num_rows();
match array.data_type() {
DataType::UInt16 => convert_array!(array, num_rows, rows, UInt16Array, Int16, i16),
DataType::Int8 => convert_array!(array, num_rows, rows, Int8Array, Int16, i16),
DataType::UInt8 => convert_array!(array, num_rows, rows, UInt8Array, Int16, i16),
DataType::Int16 => convert_array!(array, num_rows, rows, Int16Array, Int16, i16),
DataType::UInt32 => convert_array!(array, num_rows, rows, UInt32Array, Int32, i32),
DataType::UInt16 => convert_array!(array, num_rows, rows, UInt16Array, Int32, i32),
DataType::Int32 => convert_array!(array, num_rows, rows, Int32Array, Int32, i32),
DataType::UInt64 => convert_array!(array, num_rows, rows, UInt64Array, Int64, i64),
DataType::UInt32 => convert_array!(array, num_rows, rows, UInt32Array, Int64, i64),
DataType::Int64 => convert_array!(array, num_rows, rows, Int64Array, Int64, i64),
DataType::UInt64 => {
let a = array.as_any().downcast_ref::<UInt64Array>().unwrap();
for i in 0..num_rows {
rows[i].push(if a.is_null(i) {
TableValue::Null
} else {
TableValue::Decimal128(Decimal128Value::new(a.value(i) as i128, 0))
});
}
}
DataType::Boolean => {
convert_array!(array, num_rows, rows, BooleanArray, Boolean, bool)
}
Expand Down Expand Up @@ -685,7 +697,16 @@ pub fn batches_to_dataframe(

#[cfg(test)]
mod tests {
use std::sync::Arc;

use datafusion::arrow::array::PrimitiveArray;
use itertools::Itertools;

use super::*;
use crate::compile::arrow::{
datatypes::{ArrowPrimitiveType, Field},
record_batch::RecordBatchOptions,
};

#[test]
fn test_dataframe_print() {
Expand Down Expand Up @@ -815,14 +836,14 @@ mod tests {
(DataType::Float32, ColumnType::Double),
(DataType::Float64, ColumnType::Double),
(DataType::Boolean, ColumnType::Boolean),
(DataType::Int8, ColumnType::Int32),
(DataType::UInt8, ColumnType::Int32),
(DataType::Int16, ColumnType::Int32),
(DataType::UInt16, ColumnType::Int32),
(DataType::Int32, ColumnType::Int32),
(DataType::UInt32, ColumnType::Int32),
(DataType::Int8, ColumnType::Int64),
(DataType::Int16, ColumnType::Int64),
(DataType::UInt32, ColumnType::Int64),
(DataType::Int64, ColumnType::Int64),
(DataType::UInt8, ColumnType::Int64),
(DataType::UInt16, ColumnType::Int64),
(DataType::UInt64, ColumnType::Int64),
(DataType::UInt64, ColumnType::Decimal(39, 0)),
(DataType::Null, ColumnType::String),
];

Expand All @@ -831,4 +852,197 @@ mod tests {
assert_eq!(result, expected_column_type, "Failed for {:?}", arrow_type);
}
}

fn create_record_batch<T: ArrowPrimitiveType>(
data_type: DataType,
value: PrimitiveArray<T>,
expected_data_type: ColumnType,
expected_data: Vec<TableValue>,
) -> Result<(), CubeError> {
let batch = RecordBatch::try_new_with_options(
Arc::new(Schema::new(vec![Field::new("data", data_type, false)])),
vec![Arc::new(value)],
&RecordBatchOptions::default(),
)
.map_err(|e| CubeError::from(e))?;

let df = batches_to_dataframe(&batch.schema(), vec![batch.clone()])?;
let colums = df.get_columns().clone();
let data = df.data;
assert_eq!(
colums.len(),
1,
"Expecting one column in DF, but: {:?}",
colums
);
assert_eq!(expected_data_type, colums.get(0).unwrap().column_type);
assert_eq!(
data.len(),
expected_data.len(),
"Expecting {} columns in DF data, but: {:?}",
expected_data.len(),
data
);
let vec1 = data.into_iter().map(|r| r.values).flatten().collect_vec();
assert_eq!(
vec1.len(),
expected_data.len(),
"Data len {} != {}",
vec1.len(),
expected_data.len()
);
assert_eq!(vec1, expected_data);
Ok(())
}

#[test]
fn test_timestamp_conversion() -> Result<(), CubeError> {
let data_nano = vec![Some(1640995200000000000)];
create_record_batch(
DataType::Timestamp(TimeUnit::Nanosecond, None),
TimestampNanosecondArray::from(data_nano.clone()),
ColumnType::Timestamp,
data_nano
.into_iter()
.map(|e| TableValue::Timestamp(TimestampValue::new(e.unwrap(), None)))
.collect::<Vec<_>>(),
)?;

let data_micro = vec![Some(1640995200000000)];
create_record_batch(
DataType::Timestamp(TimeUnit::Microsecond, None),
TimestampMicrosecondArray::from(data_micro.clone()),
ColumnType::Timestamp,
data_micro
.into_iter()
.map(|e| TableValue::Timestamp(TimestampValue::new(e.unwrap() * 1000, None)))
.collect::<Vec<_>>(),
)?;

let data_milli = vec![Some(1640995200000)];
create_record_batch(
DataType::Timestamp(TimeUnit::Millisecond, None),
TimestampMillisecondArray::from(data_milli.clone()),
ColumnType::Timestamp,
data_milli
.into_iter()
.map(|e| TableValue::Timestamp(TimestampValue::new(e.unwrap() * 1000000, None)))
.collect::<Vec<_>>(),
)
}

#[test]
fn test_signed_conversion() -> Result<(), CubeError> {
let data8 = vec![i8::MIN, -1, 0, 1, 2, 3, 4, i8::MAX];
create_record_batch(
DataType::Int8,
Int8Array::from(data8.clone()),
ColumnType::Int32, //here we are missing TableValue::Int8 to use ColumnType::Int32
data8
.into_iter()
.map(|e| TableValue::Int16(e as i16))
.collect::<Vec<_>>(),
)?;

let data16 = vec![i16::MIN, -1, 0, 1, 2, 3, 4, i16::MAX];
create_record_batch(
DataType::Int16,
Int16Array::from(data16.clone()),
ColumnType::Int32, //here we are missing ColumnType::Int16
data16
.into_iter()
.map(|e| TableValue::Int16(e))
.collect::<Vec<_>>(),
)?;

let data32 = vec![i32::MIN, -1, 0, 1, 2, 3, 4, i32::MAX];
create_record_batch(
DataType::Int32,
Int32Array::from(data32.clone()),
ColumnType::Int32,
data32
.into_iter()
.map(|e| TableValue::Int32(e))
.collect::<Vec<_>>(),
)?;

let data64 = vec![i64::MIN, -1, 0, 1, 2, 3, 4, i64::MAX];
create_record_batch(
DataType::Int64,
Int64Array::from(data64.clone()),
ColumnType::Int64,
data64
.into_iter()
.map(|e| TableValue::Int64(e))
.collect::<Vec<_>>(),
)
}

#[test]
fn test_unsigned_conversion() -> Result<(), CubeError> {
let data8 = vec![0, 1, 2, 3, 4, u8::MAX];
create_record_batch(
DataType::UInt8,
UInt8Array::from(data8.clone()),
ColumnType::Int32, //here we are missing ColumnType::Int16
data8
.into_iter()
.map(|e| TableValue::Int16(e as i16))
.collect::<Vec<_>>(),
)?;

let data16 = vec![0, 1, 2, 3, 4, u16::MAX];
create_record_batch(
DataType::UInt16,
UInt16Array::from(data16.clone()),
ColumnType::Int32,
data16
.into_iter()
.map(|e| TableValue::Int32(e as i32))
.collect::<Vec<_>>(),
)?;

let data32 = vec![0, 1, 2, 3, 4, u32::MAX];
create_record_batch(
DataType::UInt32,
UInt32Array::from(data32.clone()),
ColumnType::Int64,
data32
.into_iter()
.map(|e| TableValue::Int64(e as i64))
.collect::<Vec<_>>(),
)?;

let data64 = vec![0, 1, 2, 3, 4, u64::MAX];
create_record_batch(
DataType::UInt64,
UInt64Array::from(data64.clone()),
ColumnType::Decimal(39, 0),
data64
.into_iter()
.map(|e| TableValue::Decimal128(Decimal128Value::new(e as i128, 0)))
.collect::<Vec<_>>(),
)
}

impl PartialEq for TableValue {
fn eq(&self, other: &Self) -> bool {
match (self, other) {
(TableValue::Null, TableValue::Null) => true,
(TableValue::String(a), TableValue::String(b)) => a == b,
(TableValue::Int16(a), TableValue::Int16(b)) => a == b,
(TableValue::Int32(a), TableValue::Int32(b)) => a == b,
(TableValue::Int64(a), TableValue::Int64(b)) => a == b,
(TableValue::Boolean(a), TableValue::Boolean(b)) => a == b,
(TableValue::Float32(a), TableValue::Float32(b)) => a == b,
(TableValue::Float64(a), TableValue::Float64(b)) => a == b,
(TableValue::List(_), TableValue::List(_)) => panic!("unsupported"),
(TableValue::Decimal128(a), TableValue::Decimal128(b)) => a == b,
(TableValue::Date(a), TableValue::Date(b)) => a == b,
(TableValue::Timestamp(a), TableValue::Timestamp(b)) => a == b,
(TableValue::Interval(_), TableValue::Interval(_)) => panic!("unsupported"),
_ => false,
}
}
}
}
Loading