Rich-T-kid commented on code in PR #23187: URL: https://github.com/apache/datafusion/pull/23187#discussion_r3540920469
########## datafusion/physical-plan/src/aggregates/group_values/multi_group_by/dictionary.rs: ########## @@ -0,0 +1,401 @@ +// 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::compute::concat; +use arrow::datatypes::{ArrowDictionaryKeyType, ArrowNativeType, DataType, Field}; +use datafusion_common::Result; +use hashbrown::HashMap; +use std::marker::PhantomData; +use std::sync::{Arc, Mutex}; + +/// Use the cache path in `vectorized_equal_to` only when the number of unique lhs group IDs +/// is at most this fraction of the total comparisons (1/10 = 10%). +const CACHE_USE_THRESHOLD: usize = 10; + +pub struct DictionaryGroupValuesColumn<K: ArrowDictionaryKeyType + Send + Sync> { + inner: Box<dyn GroupColumn>, + null_array: ArrayRef, + // Mutex is required because `vectorized_equal_to` takes `&self` (GroupColumn trait + // constraint) but needs to populate this cache on the hot path. + key_to_group_cache: Mutex<HashMap<(usize, Option<usize>), bool>>, + last_values: Option<ArrayRef>, + cached_combined: Option<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, + last_values: None, + key_to_group_cache: Mutex::new(HashMap::new()), + cached_combined: None, + _phantom: PhantomData, + } + } + + /// Returns `concat([values, null_sentinel])`, rebuilding only when `values` changes. + #[inline] + fn get_combined(&mut self, values: &ArrayRef) -> Result<ArrayRef> { + if !self + .last_values + .as_ref() + .is_some_and(|v| Arc::ptr_eq(v, values)) + { + self.cached_combined = + Some(concat(&[values.as_ref(), self.null_array.as_ref()])?); + self.last_values = Some(Arc::clone(values)); + // The incoming dictionary values buffer changed, so cached comparisons are stale. + self.key_to_group_cache.lock().unwrap().clear(); + } + Ok(Arc::clone(self.cached_combined.as_ref().unwrap())) + } + + 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)) + } + + 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(); + + // When lhs group IDs are highly repeated relative to the number of comparisons, + // caching (group_id, dict_key) -> bool avoids redundant inner comparisons. + let unique_lhs_row_count = + lhs_rows.iter().collect::<hashbrown::HashSet<_>>().len(); + if unique_lhs_row_count <= rhs_rows.len() / CACHE_USE_THRESHOLD { + let mut comparison_cache = self.key_to_group_cache.lock().unwrap(); + for (position, (&lhs_row, &rhs_row)) in + lhs_rows.iter().zip(rhs_rows).enumerate() + { + if !equal_to_results.get_bit(position) { + continue; + } + let dict_key = dict.key(rhs_row); + let is_equal = *comparison_cache + .entry((lhs_row, dict_key)) + .or_insert_with(|| match dict_key { + None => self.inner.equal_to(lhs_row, &self.null_array, 0), + Some(key) => self.inner.equal_to(lhs_row, dict_values, key), + }); Review Comment: I think the best thing to do here to to avoid any regressions. That means making this implementation a thin wrapper over an inner `GroupColumns` implementation. a follow up PR can try to optimize the low-cardinality case similar to #21765 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
