Feat: add dictionaries as a supported group column type #23187

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

Open

Rich-T-kid wants to merge 25 commits into apache:main from Rich-T-kid:rich-T-kid/dictionary-groupValuesColumn-impl

+419 −21

datafusion/physical-plan/src/aggregates/group_values/multi_group_by/dictionary.rs

-Original file line number
+Diff line change
@@ -0,0 +1,365 @@
+    // Licensed to the Apache Software Foundation (ASF) under one
+    // or more contributor license agreements.  See the NOTICE file
+    // distributed with this work for additional information
+    // regarding copyright ownership.  The ASF licenses this file
+    // to you under the Apache License, Version 2.0 (the
+    // "License"); you may not use this file except in compliance
+    // with the License.  You may obtain a copy of the License at
+    //
+    //   http://www.apache.org/licenses/LICENSE-2.0
+    //
+    // Unless required by applicable law or agreed to in writing,
+    // software distributed under the License is distributed on an
+    // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+    // KIND, either express or implied.  See the License for the
+    // specific language governing permissions and limitations
+    // under the License.
+    use crate::aggregates::group_values::multi_group_by::GroupColumn;
+    use arrow::array::{
+        Array, ArrayRef, AsArray, BooleanBufferBuilder, DictionaryArray, PrimitiveArray,
+    };
+    use arrow::datatypes::{ArrowDictionaryKeyType, ArrowNativeType, DataType, Field};
+    use datafusion_common::Result;
+    use std::marker::PhantomData;
+    use std::sync::Arc;
+    /// A [`GroupColumn`] implementation for dictionary arrays.
+    /// wraps an inner [`GroupColumn`] that stores the dictionary values, and uses a null sentinel for null keys.
+    pub struct DictionaryGroupValuesColumn<K: ArrowDictionaryKeyType + Send + Sync> {
+        inner: Box<dyn GroupColumn>,
+        // unary array used as a null sentinel for dictionary keys that are null
+        null_array: ArrayRef,
+        _phantom: PhantomData<K>,
+    }
+    impl<K: ArrowDictionaryKeyType + Send + Sync> DictionaryGroupValuesColumn<K> {
+        pub fn new(inner: Box<dyn GroupColumn>, field: &Field) -> Self {
+            let null_array = arrow::array::new_null_array(field.data_type(), 1);
+            Self {
+                inner,
+                null_array,
+                _phantom: PhantomData,
+            }
+        }
+        fn into_dict(values: ArrayRef) -> ArrayRef {
+            let num_values = values.len();
+            assert!(
+                Self::valid_bounds::<K>(num_values),
+                "Dictionary key type {:?} cannot hold {} values",
+                K::DATA_TYPE,
+                num_values
+            );
+            let keys: PrimitiveArray<K> = (0..num_values)
+                .map(|i| (!values.is_null(i)).then(|| K::Native::usize_as(i)))
+                .collect();
+            Arc::new(DictionaryArray::<K>::new(keys, values))
+        }
+        /// Returns the index of the first null in `values`, or `None` if there are no nulls.
+        fn null_sentinel_index(values: &ArrayRef) -> Option<usize> {
+            (values.null_count() > 0)
+                .then(|| (0..values.len()).find(|&i| values.is_null(i)).unwrap())
+        }
+        fn valid_bounds<T: ArrowDictionaryKeyType>(num_values: usize) -> bool {
+            let max: usize = match T::DATA_TYPE {
+                DataType::Int8 => i8::MAX as usize,
+                DataType::Int16 => i16::MAX as usize,
+                DataType::Int32 => i32::MAX as usize,
+                DataType::Int64 => i64::MAX as usize,
+                DataType::UInt8 => u8::MAX as usize,
+                DataType::UInt16 => u16::MAX as usize,
+                DataType::UInt32 => u32::MAX as usize,
+                DataType::UInt64 => usize::MAX,
+                _ => return false,
+            };
+            num_values == 0 || num_values - 1 <= max
+        }
+    }
+    impl<K: ArrowDictionaryKeyType + Send + Sync> GroupColumn
+        for DictionaryGroupValuesColumn<K>
+    {
+        fn equal_to(&self, lhs_row: usize, array: &ArrayRef, rhs_row: usize) -> bool {
+            let dict = array.as_dictionary::<K>();
+            match dict.key(rhs_row) {
+                None => self.inner.equal_to(lhs_row, &self.null_array, 0),
+                Some(key) => self.inner.equal_to(lhs_row, dict.values(), key),
+            }
+        }
+        fn append_val(&mut self, array: &ArrayRef, row: usize) -> Result<()> {
+            let dict = array.as_dictionary::<K>();
+            match dict.key(row) {
+                None => self.inner.append_val(&self.null_array, 0),
+                Some(key) => self.inner.append_val(dict.values(), key),
+            }
+        }
+        fn vectorized_equal_to(
+            &self,
+            lhs_rows: &[usize],
+            array: &ArrayRef,
+            rhs_rows: &[usize],
+            equal_to_results: &mut BooleanBufferBuilder,
+        ) {
+            let dict = array.as_dictionary::<K>();
+            let dict_values = dict.values();
+            let dict_keys = dict.keys();
+            if dict_keys.null_count() == 0 {
+                // Fast path: no nulls in the key array, resolve all indices up front and
+                // delegate to the inner column's vectorized comparison.
+                let value_indices: Vec<usize> = rhs_rows
+                    .iter()
+                    .map(|&row_index| dict_keys.value(row_index).as_usize())
+                    .collect();
+                self.inner.vectorized_equal_to(
+                    lhs_rows,
+                    dict_values,
+                    &value_indices,
+                    equal_to_results,
+                );
+            } else if let Some(sentinel) = Self::null_sentinel_index(dict_values) {
+                // Values already contains a null entry; reuse its position as the
+                // sentinel so null keys map there without any allocation.
+                let value_indices: Vec<usize> = rhs_rows
+                    .iter()
+                    .map(|&row_index| dict.key(row_index).unwrap_or(sentinel))
+                    .collect();
+                self.inner.vectorized_equal_to(
+                    lhs_rows,
+                    dict_values,
+                    &value_indices,
+                    equal_to_results,
+                );
+            } else {
+                // Null keys but values has no null entry: scalar fallback per row.
+                for (i, (&lhs_row, &rhs_row)) in
+                    lhs_rows.iter().zip(rhs_rows.iter()).enumerate()
+                {
+                    if equal_to_results.get_bit(i) {
+                        let eq = match dict.key(rhs_row) {
+                            None => self.inner.equal_to(lhs_row, &self.null_array, 0),
+                            Some(key) => self.inner.equal_to(lhs_row, dict_values, key),
+                        };
+                        if !eq {
+                            equal_to_results.set_bit(i, false);
+                        }
+                    }
+                }
+            }
+        }
+        fn vectorized_append(&mut self, array: &ArrayRef, rows: &[usize]) -> Result<()> {
+            let dict = array.as_dictionary::<K>();
+            let dict_keys = dict.keys();
+            if dict_keys.null_count() == 0 {
+                let value_indices: Vec<usize> = rows
+                    .iter()
+                    .map(|&row_index| dict_keys.value(row_index).as_usize())
+                    .collect();
+                self.inner.vectorized_append(dict.values(), &value_indices)
+            } else if let Some(sentinel) = Self::null_sentinel_index(dict.values()) {
+                // Values already contains a null entry; reuse its position as the
+                // sentinel so null keys map there without any allocation.
+                let value_indices: Vec<usize> = rows
+                    .iter()
+                    .map(|&row_index| dict.key(row_index).unwrap_or(sentinel))
+                    .collect();
+                self.inner.vectorized_append(dict.values(), &value_indices)
+            } else {
+                // Null keys but values has no null entry: scalar fallback per row.
+                for &row_index in rows {
+                    match dict.key(row_index) {
+                        None => self.inner.append_val(&self.null_array, 0)?,
+                        Some(key) => self.inner.append_val(dict.values(), key)?,
+                    }
+                }
+                Ok(())
+            }
+        }
+        fn len(&self) -> usize {
+            self.inner.len()
+        }
+        fn size(&self) -> usize {
+            self.inner.size()
+        }
+        fn build(self: Box<Self>) -> ArrayRef {
+            Self::into_dict(self.inner.build())
+        }
+        fn take_n(&mut self, n: usize) -> ArrayRef {
+            Self::into_dict(self.inner.take_n(n))
+        }
+    }
+    #[cfg(test)]
+    mod tests {
+        use super::*;
+        use crate::aggregates::group_values::multi_group_by::bytes::ByteGroupValueBuilder;
+        use arrow::array::{
+            Array, ArrayRef, BooleanBufferBuilder, DictionaryArray, Int32Array, StringArray,
+        };
+        use arrow::compute::cast;
+        use arrow::datatypes::{DataType, Int32Type};
+        use datafusion_physical_expr::binary_map::OutputType;
+        use std::sync::Arc;
+        fn utf8_col() -> DictionaryGroupValuesColumn<Int32Type> {
+            let field = Field::new("", DataType::Utf8, true);
+            DictionaryGroupValuesColumn::<Int32Type>::new(
+                Box::new(ByteGroupValueBuilder::<i32>::new(OutputType::Utf8)),
+                &field,
+            )
+        }
+        fn dict_arr(keys: &[Option<i32>], values: &[&str]) -> ArrayRef {
+            Arc::new(DictionaryArray::<Int32Type>::new(
+                Int32Array::from(keys.to_vec()),
+                Arc::new(StringArray::from(values.to_vec())),
+            ))
+        }
+        fn str_values(arr: &ArrayRef) -> Vec<Option<String>> {
+            let plain = cast(arr.as_ref(), &DataType::Utf8).unwrap();
+            let strings = plain.as_any().downcast_ref::<StringArray>().unwrap();
+            (0..strings.len())
+                .map(|i| strings.is_valid(i).then(|| strings.value(i).to_owned()))
+                .collect()
+        }
+        fn assert_is_dict_utf8(arr: &ArrayRef) {
+            assert!(
+                matches!(arr.data_type(),
+                    DataType::Dictionary(k, v)
+                        if k.as_ref() == &DataType::Int32 && v.as_ref() == &DataType::Utf8),
+                "expected Dictionary(Int32, Utf8), got {:?}",
+                arr.data_type()
+            );
+        }
+        fn true_buf(len: usize) -> BooleanBufferBuilder {
+            let mut buf = BooleanBufferBuilder::new(len);
+            buf.append_n(len, true);
+            buf
+        }
+        fn buf_to_vec(buf: &BooleanBufferBuilder) -> Vec<bool> {
+            (0..buf.len()).map(|i| buf.get_bit(i)).collect()
+        }
+        #[test]
+        fn utf8_dict_null_handling() {
+            let mut col = utf8_col();
+            let input: ArrayRef = Arc::new(DictionaryArray::<Int32Type>::new(
+                Int32Array::from(vec![None, Some(0), Some(1)]),
+                Arc::new(StringArray::from(vec![None::<&str>, Some("b")])),
+            ));
+            for row in 0..3 {
+                col.append_val(&input, row).unwrap();
+            }
+            assert!(col.equal_to(0, &input, 0) && col.equal_to(1, &input, 1));
+            assert!(col.equal_to(0, &input, 1)); // null == null
+            assert!(!col.equal_to(0, &input, 2) && !col.equal_to(2, &input, 0));
+            let mut buf = true_buf(3);
+            col.vectorized_equal_to(&[0, 1, 2], &input, &[2, 2, 1], &mut buf);
+            assert_eq!(buf_to_vec(&buf), vec![false, false, false]);
+            let mut buf = true_buf(3);
+            buf.set_bit(0, false);
+            col.vectorized_equal_to(&[0, 1, 2], &input, &[0, 1, 2], &mut buf);
+            assert_eq!(buf_to_vec(&buf), vec![false, true, true]);
+            assert_eq!(col.take_n(0).len(), 0);
+            let taken = col.take_n(3);
+            assert_is_dict_utf8(&taken);
+            assert_eq!(str_values(&taken), vec![None, None, Some("b".into())]);
+            let input2 = dict_arr(&[Some(0), None, Some(1), Some(0), None], &["cat", "dog"]);
+            col.vectorized_append(&input2, &[0, 1, 2, 3, 4]).unwrap();
+            let taken2 = col.take_n(5);
+            assert_is_dict_utf8(&taken2);
+            assert_eq!(
+                str_values(&taken2),
+                vec![
+                    Some("cat".into()),
+                    None,
+                    Some("dog".into()),
+                    Some("cat".into()),
+                    None,
+                ]
+            );
+            let empty = Box::new(col).build();
+            assert_is_dict_utf8(&empty);
+            assert_eq!(empty.len(), 0);
+        }
+        #[test]
+        fn u64_primitive_inner_multi_batch() {
+            use crate::aggregates::group_values::multi_group_by::primitive::PrimitiveGroupValueBuilder;
+            use arrow::array::UInt64Array;
+            use arrow::datatypes::UInt64Type;
+            let field = Field::new("", DataType::UInt64, true);
+            let mut col = DictionaryGroupValuesColumn::<Int32Type>::new(
+                Box::new(PrimitiveGroupValueBuilder::<UInt64Type, true>::new(
+                    DataType::UInt64,
+                )),
+                &field,
+            );
+            let batch1: ArrayRef = Arc::new(DictionaryArray::<Int32Type>::new(
+                Int32Array::from(vec![Some(0), Some(1), None, Some(2), Some(0)]),
+                Arc::new(UInt64Array::from(vec![10u64, 20, 30])),
+            ));
+            col.vectorized_append(&batch1, &[0, 1, 2, 3, 4]).unwrap();
+            let mut buf = true_buf(5);
+            col.vectorized_equal_to(&[0, 1, 2, 3, 4], &batch1, &[0, 1, 2, 3, 0], &mut buf);
+            assert_eq!(buf_to_vec(&buf), vec![true, true, true, true, true]);
+            let mut buf = true_buf(2);
+            col.vectorized_equal_to(&[0, 4], &batch1, &[1, 1], &mut buf);
+            assert_eq!(buf_to_vec(&buf), vec![false, false]);
+            let u64_val = |arr: &ArrayRef, pos: usize| {
+                let dict = arr.as_dictionary::<Int32Type>();
+                dict.values()
+                    .as_any()
+                    .downcast_ref::<UInt64Array>()
+                    .unwrap()
+                    .value(dict.key(pos).unwrap())
+            };
+            let taken1 = col.take_n(3);
+            assert_eq!(u64_val(&taken1, 0), 10);
+            assert_eq!(u64_val(&taken1, 1), 20);
+            assert!(taken1.as_dictionary::<Int32Type>().key(2).is_none());
+            let batch2: ArrayRef = Arc::new(DictionaryArray::<Int32Type>::new(
+                Int32Array::from(vec![None, Some(0)]),
+                Arc::new(UInt64Array::from(vec![99u64])),
+            ));
+            col.vectorized_append(&batch2, &[0, 1]).unwrap();
+            let mut buf = true_buf(2);
+            col.vectorized_equal_to(&[2, 3], &batch2, &[0, 1], &mut buf);
+            assert_eq!(buf_to_vec(&buf), vec![true, true]);
+            let out = Box::new(col).build();
+            assert_eq!(u64_val(&out, 0), 30);
+            assert_eq!(u64_val(&out, 1), 10);
+            assert!(out.as_dictionary::<Int32Type>().key(2).is_none());
+            assert_eq!(u64_val(&out, 3), 99);
+        }
+    }

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Feat: add dictionaries as a supported group column type #23187

Diff view

Diff view

There are no files selected for viewing

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Feat: add dictionaries as a supported group column type #23187

Are you sure you want to change the base?

Feat: add dictionaries as a supported group column type #23187

Uh oh!

Uh oh!

Diff view

Diff view

There are no files selected for viewing

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!