Rich-T-kid commented on code in PR #22859: URL: https://github.com/apache/datafusion/pull/22859#discussion_r3414036211
########## datafusion/physical-plan/benches/multi_column_dictionary_group_values.rs: ########## @@ -0,0 +1,218 @@ +// 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. + +//! Benchmarks for `GroupValues` over multiple `Dictionary<Int32, Utf8>` columns. +//! Covers 4 and 8 group-by columns, batch sizes of 8 KiB and 64 KiB rows, +//! and cardinalities realistic for multi-column GROUP BY workloads (20 / 100 / 500 / 1 000). + +use arrow::array::{ArrayRef, DictionaryArray, PrimitiveArray, StringArray}; +use arrow::buffer::{Buffer, NullBuffer}; +use arrow::datatypes::{DataType, Field, Schema, SchemaRef, UInt64Type}; +use criterion::{ + BatchSize, BenchmarkId, Criterion, Throughput, criterion_group, criterion_main, +}; +use datafusion_expr::EmitTo; +use datafusion_physical_plan::aggregates::group_values::new_group_values; +use datafusion_physical_plan::aggregates::order::GroupOrdering; +use rand::rngs::StdRng; +use rand::seq::SliceRandom; +use rand::{Rng, SeedableRng}; +use std::hint::black_box; +use std::sync::Arc; + +const SIZES: [usize; 2] = [8 * 1024, 64 * 1024]; +const N_COLS: [usize; 2] = [4, 8]; +const CARDS: [usize; 4] = [20, 100, 500, 1_000]; +const N_BATCHES: usize = 5; +const NULL_DENSITY: f32 = 0.15; +const SEED: u64 = 0xD1C7; +/// Per-column cardinality variance as a percentage of the target (half above, half below). +/// each column's distinct count is sampled from [target*0.95, target*1.05]. +const CARDINALITY_RANGE: usize = 10; + +fn schema_for_cols(n_cols: usize) -> SchemaRef { + let dict_ty = + DataType::Dictionary(Box::new(DataType::UInt64), Box::new(DataType::Utf8)); + let fields: Vec<Field> = (0..n_cols) + .map(|i| Field::new(format!("g{i}"), dict_ty.clone(), true)) + .collect(); + Arc::new(Schema::new(fields)) +} + +fn make_dict_col( + size: usize, + num_distinct: usize, + null_density: f32, + seed: u64, +) -> ArrayRef { + let mut rng = StdRng::seed_from_u64(seed); + + let strings: Vec<String> = (0..num_distinct) + .map(|i| format!("dict_label_{i:012}")) + .collect(); + let values = Arc::new(StringArray::from( + strings.iter().map(String::as_str).collect::<Vec<_>>(), + )); + + // When the pool is at least as large as the batch, shuffle a prefix so + // every row in this batch maps to a distinct key. + let keys: Vec<u64> = if num_distinct >= size { + let mut perm: Vec<u64> = (0..size as u64).collect(); + perm.shuffle(&mut rng); + perm + } else { + (0..size) Review Comment: good catch! previously column_count * distinct was being generated. I've updated the code to generate the group id's first and then pass it into the creation of the dictionarys. this way no matter how many columns get produced they will produce the same number of group id's. Ive also added benchmarks for the cross-product group-bys to try and better reflect the worse case scenario. ex ``` group by department, city,rank ``` -- 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]
