sunchao commented on code in PR #7173: URL: https://github.com/apache/datafusion-comet/pull/7173#discussion_r4365203392
########## native/spark-expr/tests/bloom_filter_spill.rs: ########## @@ -0,0 +1,160 @@ +// 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 std::sync::Arc; + +use arrow::array::{Array, BinaryArray, BinaryBuilder, Int64Array, RecordBatch}; +use arrow::datatypes::{DataType, Field, Schema}; +use datafusion::common::{Result, ScalarValue}; +use datafusion::datasource::memory::MemorySourceConfig; +use datafusion::datasource::source::DataSourceExec; +use datafusion::execution::memory_pool::{GreedyMemoryPool, MemoryPool}; +use datafusion::execution::runtime_env::RuntimeEnvBuilder; +use datafusion::execution::TaskContext; +use datafusion::logical_expr::AggregateUDF; +use datafusion::physical_expr::aggregate::{AggregateExprBuilder, AggregateFunctionExpr}; +use datafusion::physical_expr::expressions::{Column, Literal}; +use datafusion::physical_plan::aggregates::{AggregateExec, AggregateMode, PhysicalGroupBy}; +use datafusion::physical_plan::{common::collect, ExecutionPlan}; +use datafusion::prelude::SessionConfig; +use datafusion_comet_spark_expr::{BloomFilterAgg, SparkBloomFilterVersion}; + +fn bloom_state(aggregate: &AggregateFunctionExpr, values: &[i64]) -> Result<Vec<u8>> { + let mut accumulator = aggregate.create_accumulator()?; + accumulator.update_batch(&[Arc::new(Int64Array::from_iter_values( + values.iter().copied(), + ))])?; + let ScalarValue::Binary(Some(state)) = accumulator.state()?.remove(0) else { + panic!("a nonempty Bloom filter must have a binary state"); + }; + Ok(state) +} + +#[tokio::test] +async fn grouped_bloom_filter_merges_spilled_states() -> Result<()> { + const NUM_GROUPS: usize = 1024; + const BATCH_SIZE: usize = 8; + const NUM_BITS: i64 = 131_072; + + let value_schema = Arc::new(Schema::new(vec![ + Field::new("key", DataType::Int64, false), + Field::new("value", DataType::Int64, false), + ])); + let aggregate_udf = Arc::new(AggregateUDF::new_from_impl(BloomFilterAgg::new( + Arc::new(Literal::new(ScalarValue::Int64(Some(1024)))), + Arc::new(Literal::new(ScalarValue::Int64(Some(NUM_BITS)))), + DataType::Binary, + SparkBloomFilterVersion::V1, + ))); + let aggregate_expr = Arc::new( + AggregateExprBuilder::new(aggregate_udf, vec![Arc::new(Column::new("value", 1))]) + .schema(Arc::clone(&value_schema)) + .alias("bloom") + .build()?, + ); + let state_schema = Arc::new(Schema::new(vec![ + Field::new("key", DataType::Int64, false), + Field::new("bits", DataType::Binary, false), + ])); + + // Each pass has one valid, nonempty partial state per group. The complete set + // of 16 KiB filters exceeds the pool, so the two passes meet during sorted + // spill replay. Distinct values detect lost states as well as merge panics. + let mut batches = Vec::new(); + for pass in 0..2 { + for start in (0..NUM_GROUPS).step_by(BATCH_SIZE) { + let mut states = BinaryBuilder::new(); + for key in start..start + BATCH_SIZE { + states.append_value(bloom_state( + &aggregate_expr, + &[(pass * NUM_GROUPS + key) as i64], + )?); + } + batches.push(RecordBatch::try_new( + Arc::clone(&state_schema), + vec![ + Arc::new(Int64Array::from_iter_values( + (start..start + BATCH_SIZE).map(|key| key as i64), + )), + Arc::new(states.finish()), + ], + )?); + } + } + let scan = Arc::new(DataSourceExec::new(Arc::new(MemorySourceConfig::try_new( + &[batches], + state_schema, + None, + )?))); + let aggregate = AggregateExec::try_new( + AggregateMode::Final, + PhysicalGroupBy::new_single(vec![(Arc::new(Column::new("key", 0)), "key".into())]), + vec![Arc::clone(&aggregate_expr)], + vec![None], + scan, + value_schema, + )?; + let pool = Arc::new(GreedyMemoryPool::new(8 * 1024 * 1024)); + let runtime = RuntimeEnvBuilder::new() + .with_memory_pool(Arc::clone(&pool) as _) + .build_arc()?; + let context = Arc::new( + TaskContext::default() + .with_runtime(runtime) + .with_session_config( + // Exercise the sorted-spill replay path explicitly. + SessionConfig::new() + .with_batch_size(BATCH_SIZE) + .set_bool("datafusion.execution.enable_migration_aggregate", false), Review Comment: I replaced the forced-spill fixture in 9a83585f6 with a grouped `Final` aggregate using DataFusion’s default settings. One input batch gives group 1 partial filters for 11 and 22 and group 2 a filter for 99. The test checks the exact bytes for both groups. It passed in the focused harness and failed against the original accumulator at the one-row assertion (`2` rows received, `1` expected). -- 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]
