westonpace commented on a change in pull request #12537: URL: https://github.com/apache/arrow/pull/12537#discussion_r836846192
########## File path: cpp/src/arrow/compute/exec/tpch_benchmark.cc ########## @@ -0,0 +1,123 @@ +// 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. + +#include <benchmark/benchmark.h> + +#include "arrow/compute/cast.h" +#include "arrow/compute/exec/test_util.h" +#include "arrow/compute/exec/tpch_node.h" +#include "arrow/testing/future_util.h" +#include "arrow/util/make_unique.h" + +namespace arrow { +namespace compute { +namespace internal { + +std::shared_ptr<ExecPlan> Plan_Q1(AsyncGenerator<util::optional<ExecBatch>>* sink_gen, + int scale_factor) { + ExecContext* ctx = default_exec_context(); + *ctx = ExecContext(default_memory_pool(), arrow::internal::GetCpuThreadPool()); + std::shared_ptr<ExecPlan> plan = *ExecPlan::Make(ctx); + TpchGen gen = *TpchGen::Make(plan.get(), static_cast<double>(scale_factor)); + + ExecNode* lineitem = + *gen.Lineitem({"L_QUANTITY", "L_EXTENDEDPRICE", "L_TAX", "L_DISCOUNT", "L_SHIPDATE", + "L_RETURNFLAG", "L_LINESTATUS"}); + + auto sept_2_1998 = std::make_shared<Date32Scalar>( + 10471); // September 2, 1998 is 10471 days after January 1, 1970 + Expression filter = + less_equal(field_ref("L_SHIPDATE"), literal(std::move(sept_2_1998))); + FilterNodeOptions filter_opts(filter); + + Expression l_returnflag = field_ref("L_RETURNFLAG"); + Expression l_linestatus = field_ref("L_LINESTATUS"); + Expression quantity = field_ref("L_QUANTITY"); + Expression base_price = field_ref("L_EXTENDEDPRICE"); + + std::shared_ptr<Decimal128Scalar> decimal_1 = + std::make_shared<Decimal128Scalar>(Decimal128{0, 100}, decimal(12, 2)); + Expression discount_multiplier = + call("subtract", {literal(decimal_1), field_ref("L_DISCOUNT")}); + Expression tax_multiplier = call("add", {literal(decimal_1), field_ref("L_TAX")}); + Expression disc_price = + call("multiply", {field_ref("L_EXTENDEDPRICE"), discount_multiplier}); + Expression charge = + call("multiply", + {call("cast", + {call("multiply", {field_ref("L_EXTENDEDPRICE"), discount_multiplier})}, + compute::CastOptions::Unsafe(decimal(12, 2))), + tax_multiplier}); + Expression discount = field_ref("L_DISCOUNT"); + + std::vector<Expression> projection_list = {l_returnflag, l_linestatus, quantity, + base_price, disc_price, charge, + quantity, base_price, discount}; + std::vector<std::string> project_names = { + "l_returnflag", "l_linestatus", "sum_qty", "sum_base_price", "sum_disc_price", + "sum_charge", "avg_qty", "avg_price", "avg_disc"}; + ProjectNodeOptions project_opts(std::move(projection_list), std::move(project_names)); + + ScalarAggregateOptions sum_opts = ScalarAggregateOptions::Defaults(); + CountOptions count_opts(CountOptions::CountMode::ALL); + std::vector<arrow::compute::internal::Aggregate> aggs = { + {"hash_sum", &sum_opts}, {"hash_sum", &sum_opts}, {"hash_sum", &sum_opts}, + {"hash_sum", &sum_opts}, {"hash_mean", &sum_opts}, {"hash_mean", &sum_opts}, + {"hash_mean", &sum_opts}, {"hash_count", &count_opts}}; + + std::vector<FieldRef> to_aggregate = {"sum_qty", "sum_base_price", "sum_disc_price", + "sum_charge", "avg_qty", "avg_price", + "avg_disc", "sum_qty"}; + + std::vector<std::string> names = {"sum_qty", "sum_base_price", "sum_disc_price", + "sum_charge", "avg_qty", "avg_price", + "avg_disc", "count_order"}; + + std::vector<FieldRef> keys = {"l_returnflag", "l_linestatus"}; + AggregateNodeOptions agg_opts(aggs, to_aggregate, names, keys); + + SortKey l_returnflag_key("l_returnflag"); + SortKey l_linestatus_key("l_linestatus"); + SortOptions sort_opts({l_returnflag_key, l_linestatus_key}); + OrderBySinkNodeOptions order_by_opts(sort_opts, sink_gen); + + Declaration filter_decl("filter", {Declaration::Input(lineitem)}, filter_opts); + Declaration project_decl("project", project_opts); + Declaration aggregate_decl("aggregate", agg_opts); + Declaration orderby_decl("order_by_sink", order_by_opts); + + Declaration q1 = + Declaration::Sequence({filter_decl, project_decl, aggregate_decl, orderby_decl}); + std::ignore = *q1.AddToPlan(plan.get()); + return plan; +} + +static void BM_Tpch_Q1(benchmark::State& st) { + for (auto _ : st) { + st.PauseTiming(); + AsyncGenerator<util::optional<ExecBatch>> sink_gen; + std::shared_ptr<ExecPlan> plan = Plan_Q1(&sink_gen, static_cast<int>(st.range(0))); + st.ResumeTiming(); + auto fut = StartAndCollect(plan.get(), sink_gen); + auto res = *fut.MoveResult(); + } +} Review comment: I'm not sure the benchmarks will be entirely deleted. The separate tool is for profiling and not benchmarking. Ideally we can have one "builtin queries" spot where these queries are defined and then we can use them in both the query tester (for profiling) and the benchmarks (e.g. for detecting regressions). I agree that output items per second is probably meaningless. In the long term I think we would want something like "(input rows * cost_factor) per second" where `cost_factor` is some kind of empirically calculated estimate based on how long a query "should" take based on the complexity. E.g. a scan-filter-project-sink would have a relatively low cost factor but a query with joins would have a higher cost factor. That's somewhat theoretical though. Until then I think a relatively intuitive and meaningful metric would be "input bytes per second". It may not be a simple thing to add though since the size of the input is somewhat random and so we'd need to keep some kind of counters while running. Let's add a follow-up PR to address this more thoroughly. It can just be "add TPC-H benchmarks to the benchmarks suite". -- 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]
