thinkharderdev commented on code in PR #3769: URL: https://github.com/apache/arrow-datafusion/pull/3769#discussion_r992727637
########## benchmarks/src/bin/parquet_filter_pushdown.rs: ########## @@ -0,0 +1,462 @@ +// 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 arrow::array::{ + Int32Builder, StringBuilder, StringDictionaryBuilder, TimestampNanosecondBuilder, + UInt16Builder, +}; +use arrow::datatypes::{DataType, Field, Int32Type, Schema, SchemaRef, TimeUnit}; +use arrow::record_batch::RecordBatch; +use arrow::util::pretty; +use datafusion::common::Result; +use datafusion::datasource::listing::{ListingTableUrl, PartitionedFile}; +use datafusion::datasource::object_store::ObjectStoreUrl; +use datafusion::execution::context::ExecutionProps; +use datafusion::logical_expr::{lit, or, Expr}; +use datafusion::logical_plan::ToDFSchema; +use datafusion::physical_expr::create_physical_expr; +use datafusion::physical_plan::collect; +use datafusion::physical_plan::file_format::{ + FileScanConfig, ParquetExec, ParquetScanOptions, +}; +use datafusion::physical_plan::filter::FilterExec; +use datafusion::prelude::{col, combine_filters, SessionConfig, SessionContext}; +use object_store::path::Path; +use object_store::ObjectMeta; +use parquet::arrow::ArrowWriter; +use rand::rngs::StdRng; +use rand::{Rng, SeedableRng}; +use std::fs::File; +use std::ops::Range; +use std::path::PathBuf; +use std::sync::Arc; +use std::time::Instant; +use structopt::StructOpt; + +#[cfg(feature = "snmalloc")] +#[global_allocator] +static ALLOC: snmalloc_rs::SnMalloc = snmalloc_rs::SnMalloc; + +#[derive(Debug, StructOpt)] +#[structopt(name = "Benchmarks", about = "Apache Arrow Rust Benchmarks.")] +struct Opt { + /// Activate debug mode to see query results + #[structopt(short, long)] + debug: bool, + + /// Number of iterations of each test run + #[structopt(short = "i", long = "iterations", default_value = "3")] + iterations: usize, + + /// Number of partitions to process in parallel + #[structopt(long = "partitions", default_value = "2")] + partitions: usize, + + /// Path to folder where access log file will be generated + #[structopt(parse(from_os_str), required = true, short = "p", long = "path")] + path: PathBuf, + + /// Batch size when reading Parquet files + #[structopt(short = "s", long = "batch-size", default_value = "8192")] + batch_size: usize, + + /// Total size of generated dataset. The default scale factor of 1.0 will generate a roughly 1GB parquet file + #[structopt(short = "s", long = "scale-factor", default_value = "1.0")] + scale_factor: f32, +} + +#[tokio::main] +async fn main() -> Result<()> { + let opt: Opt = Opt::from_args(); + println!("Running benchmarks with the following options: {:?}", opt); + + let config = SessionConfig::new() + .with_target_partitions(opt.partitions) + .with_batch_size(opt.batch_size); + let mut ctx = SessionContext::with_config(config); + + let path = opt.path.join("logs.parquet"); + + let (object_store_url, object_meta) = gen_data(path, opt.scale_factor)?; + + run_benchmarks( + &mut ctx, + object_store_url.clone(), + object_meta.clone(), + opt.iterations, + opt.debug, + ) + .await?; + + Ok(()) +} + +async fn run_benchmarks( + ctx: &mut SessionContext, + object_store_url: ObjectStoreUrl, + object_meta: ObjectMeta, + iterations: usize, + debug: bool, +) -> Result<()> { + let scan_options_matrix = vec![ + ParquetScanOptions::default(), + ParquetScanOptions::default() + .with_page_index(true) + .with_pushdown_filters(true) + .with_reorder_predicates(true), + ParquetScanOptions::default() + .with_page_index(true) + .with_pushdown_filters(true) + .with_reorder_predicates(false), + ]; + + let filter_matrix = vec![ + // Selective-ish filter + col("request_method").eq(lit("GET")), + // Non-selective filter + col("request_method").not_eq(lit("GET")), + // Basic conjunction + col("request_method") + .eq(lit("POST")) + .and(col("response_status").eq(lit(503_u16))), + // Nested filters + col("request_method").eq(lit("POST")).and(or( + col("response_status").eq(lit(503_u16)), + col("response_status").eq(lit(403_u16)), + )), + // Many filters + combine_filters(&[ + col("request_method").not_eq(lit("GET")), + col("response_status").eq(lit(400_u16)), + // TODO this fails in the FilterExec with Error: Internal("The type of Dictionary(Int32, Utf8) = Utf8 of binary physical should be same") + // col("service").eq(lit("backend")), + ]) + .unwrap(), + // Filter everything + col("response_status").eq(lit(429_u16)), + // Filter nothing + col("response_status").gt(lit(0_u16)), + ]; + + for filter_expr in &filter_matrix { + println!("Executing with filter '{}'", filter_expr); + for scan_options in &scan_options_matrix { + println!("Using scan options {:?}", scan_options); + for i in 0..iterations { + let start = Instant::now(); + let rows = exec_scan( + ctx, + object_store_url.clone(), + object_meta.clone(), + filter_expr.clone(), + scan_options.clone(), + debug, + ) + .await?; + println!( + "Iteration {} returned {} rows in {} ms", + i, + rows, + start.elapsed().as_millis() + ); + } + } + println!("\n"); + } + Ok(()) +} + +async fn exec_scan( + ctx: &SessionContext, + object_store_url: ObjectStoreUrl, + object_meta: ObjectMeta, + filter: Expr, + scan_options: ParquetScanOptions, + debug: bool, +) -> Result<usize> { + let schema = BatchBuilder::schema(); + let scan_config = FileScanConfig { + object_store_url, + file_schema: schema.clone(), + file_groups: vec![vec![PartitionedFile { + object_meta, + partition_values: vec![], + range: None, + extensions: None, + }]], + statistics: Default::default(), + projection: None, + limit: None, + table_partition_cols: vec![], + }; + + let df_schema = schema.clone().to_dfschema()?; + + let physical_filter_expr = create_physical_expr( + &filter, + &df_schema, + schema.as_ref(), + &ExecutionProps::default(), + )?; + + let parquet_exec = Arc::new( + ParquetExec::new(scan_config, Some(filter), None).with_scan_options(scan_options), + ); + + let exec = Arc::new(FilterExec::try_new(physical_filter_expr, parquet_exec)?); + + let task_ctx = ctx.task_ctx(); + let result = collect(exec, task_ctx).await?; + + if debug { + pretty::print_batches(&result)?; + } + Ok(result.iter().map(|b| b.num_rows()).sum()) +} + +fn gen_data(path: PathBuf, scale_factor: f32) -> Result<(ObjectStoreUrl, ObjectMeta)> { + let generator = Generator::new(); + + let file = File::create(&path).unwrap(); + let mut writer = ArrowWriter::try_new(file, generator.schema.clone(), None).unwrap(); Review Comment: Yeah, I'll need to revisit this again once https://github.com/apache/arrow-rs/pull/2854 is released and pulled in so we cam generate the files with proper page sizes (which should make a significant difference) -- 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]
