westonpace commented on code in PR #13366: URL: https://github.com/apache/arrow/pull/13366#discussion_r899691871
########## cpp/src/arrow/compute/exec/filter_benchmark.cc: ########## @@ -0,0 +1,176 @@ +// 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 <condition_variable> +#include <mutex> + +#include "benchmark/benchmark.h" + +#include "arrow/compute/cast.h" +#include "arrow/compute/exec.h" +#include "arrow/compute/exec/expression.h" +#include "arrow/compute/exec/options.h" +#include "arrow/compute/exec/task_util.h" +#include "arrow/compute/exec/test_util.h" +#include "arrow/dataset/partition.h" +#include "arrow/testing/future_util.h" +#include "arrow/testing/generator.h" +#include "arrow/testing/gtest_util.h" +#include "arrow/type.h" + +namespace arrow { +namespace compute { + +static constexpr int64_t kTotalBatchSize = 1000000; + +static void FilterOverhead(benchmark::State& state, Expression expr) { + const int32_t batch_size = static_cast<int32_t>(state.range(0)); + const int32_t num_batches = kTotalBatchSize / batch_size; + + arrow::compute::BatchesWithSchema data = MakeRandomBatches( + schema({field("i64", int64()), field("bool", boolean())}), num_batches, batch_size); + ExecContext ctx(default_memory_pool(), arrow::internal::GetCpuThreadPool()); + for (auto _ : state) { + state.PauseTiming(); + ASSERT_OK_AND_ASSIGN(std::shared_ptr<arrow::compute::ExecPlan> plan, + ExecPlan::Make(&ctx)); + AsyncGenerator<util::optional<ExecBatch>> sink_gen; + ASSERT_OK(Declaration::Sequence( + { + {"source", + SourceNodeOptions{data.schema, + data.gen(/*parallel=*/true, /*slow=*/false)}, + "custom_source_label"}, + {"filter", + FilterNodeOptions{ + expr, + }}, + {"sink", SinkNodeOptions{&sink_gen}, "custom_sink_label"}, + }) + .AddToPlan(plan.get())); + state.ResumeTiming(); + ASSERT_FINISHES_OK(StartAndCollect(plan.get(), sink_gen)); + } + state.counters["rows_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches * batch_size), + benchmark::Counter::kIsRate); + + state.counters["batches_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches), benchmark::Counter::kIsRate); +} + +static void FilterOverheadIsolated(benchmark::State& state, Expression expr) { + const int32_t batch_size = static_cast<int32_t>(state.range(0)); + const int32_t num_batches = kTotalBatchSize / batch_size; + + arrow::compute::BatchesWithSchema data = MakeRandomBatches( + schema({field("i64", int64()), field("bool", boolean())}), num_batches, batch_size); + ExecContext ctx(default_memory_pool(), arrow::internal::GetCpuThreadPool()); + for (auto _ : state) { + state.PauseTiming(); + AsyncGenerator<util::optional<ExecBatch>> sink_gen; + + ASSERT_OK_AND_ASSIGN(std::shared_ptr<arrow::compute::ExecPlan> plan, + ExecPlan::Make(&ctx)); + // Source and sink nodes have no effect on the benchmark. + // Used for dummy purposes as they are referenced in InputReceived and InputFinished. + ASSERT_OK_AND_ASSIGN( + arrow::compute::ExecNode * source_node, + MakeExecNode("source", plan.get(), {}, + SourceNodeOptions{data.schema, data.gen(/*parallel=*/true, + /*slow=*/false)})); + ASSERT_OK_AND_ASSIGN(arrow::compute::ExecNode * project_node, + MakeExecNode("filter", plan.get(), {source_node}, + FilterNodeOptions{ + expr, + })); + MakeExecNode("sink", plan.get(), {project_node}, SinkNodeOptions{&sink_gen}); + + std::unique_ptr<arrow::compute::TaskScheduler> scheduler = TaskScheduler::Make(); + std::condition_variable all_tasks_finished_cv; + std::mutex mutex; + int task_group_id = scheduler->RegisterTaskGroup( + [&](size_t thread_id, int64_t task_id) { + project_node->InputReceived(source_node, data.batches[task_id]); + return Status::OK(); + }, + [&](size_t thread_id) { + project_node->InputFinished(source_node, static_cast<int>(data.batches.size())); + std::unique_lock<std::mutex> lk(mutex); + all_tasks_finished_cv.notify_one(); + return Status::OK(); + }); + scheduler->RegisterEnd(); + ThreadIndexer thread_indexer; + + state.ResumeTiming(); + arrow::internal::ThreadPool* thread_pool = arrow::internal::GetCpuThreadPool(); + ASSERT_OK(scheduler->StartScheduling( + thread_indexer(), + [&](std::function<Status(size_t)> task) -> Status { + return thread_pool->Spawn([&, task]() { + size_t tid = thread_indexer(); + ARROW_DCHECK_OK(task(tid)); + }); + }, + thread_pool->GetCapacity(), + /*use_sync_execution=*/false)); + std::unique_lock<std::mutex> lk(mutex); + ASSERT_OK(scheduler->StartTaskGroup(thread_indexer(), task_group_id, num_batches)); + all_tasks_finished_cv.wait(lk); + ASSERT_TRUE(project_node->finished().is_finished()); + } + state.counters["rows_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches * batch_size), + benchmark::Counter::kIsRate); + + state.counters["batches_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches), benchmark::Counter::kIsRate); +} + +arrow::compute::Expression complex_expression = + less(less(field_ref("i64"), literal(20)), greater(field_ref("i64"), literal(0))); +arrow::compute::Expression simple_expression = + less(call("negate", {field_ref("i64")}), literal(0)); +arrow::compute::Expression zero_copy_expression = is_valid((call( + "cast", {field_ref("i64")}, compute::CastOptions::Safe(timestamp(TimeUnit::NANO))))); +arrow::compute::Expression ref_only_expression = less(field_ref("i64"), literal(0)); Review Comment: I think that's fine. I'm not sure `ref_only_expression` is the same thing though? Perhaps `ref_only_expressions` could be something like `field_ref("bool")` where `bool` is a randomly generated boolean array? ########## cpp/src/arrow/compute/exec/filter_benchmark.cc: ########## @@ -0,0 +1,176 @@ +// 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 <condition_variable> +#include <mutex> + +#include "benchmark/benchmark.h" + +#include "arrow/compute/cast.h" +#include "arrow/compute/exec.h" +#include "arrow/compute/exec/expression.h" +#include "arrow/compute/exec/options.h" +#include "arrow/compute/exec/task_util.h" +#include "arrow/compute/exec/test_util.h" +#include "arrow/dataset/partition.h" +#include "arrow/testing/future_util.h" +#include "arrow/testing/generator.h" +#include "arrow/testing/gtest_util.h" +#include "arrow/type.h" + +namespace arrow { +namespace compute { + +static constexpr int64_t kTotalBatchSize = 1000000; + +static void FilterOverhead(benchmark::State& state, Expression expr) { + const int32_t batch_size = static_cast<int32_t>(state.range(0)); + const int32_t num_batches = kTotalBatchSize / batch_size; + + arrow::compute::BatchesWithSchema data = MakeRandomBatches( + schema({field("i64", int64()), field("bool", boolean())}), num_batches, batch_size); + ExecContext ctx(default_memory_pool(), arrow::internal::GetCpuThreadPool()); + for (auto _ : state) { + state.PauseTiming(); + ASSERT_OK_AND_ASSIGN(std::shared_ptr<arrow::compute::ExecPlan> plan, + ExecPlan::Make(&ctx)); + AsyncGenerator<util::optional<ExecBatch>> sink_gen; + ASSERT_OK(Declaration::Sequence( + { + {"source", + SourceNodeOptions{data.schema, + data.gen(/*parallel=*/true, /*slow=*/false)}, + "custom_source_label"}, + {"filter", + FilterNodeOptions{ + expr, + }}, + {"sink", SinkNodeOptions{&sink_gen}, "custom_sink_label"}, + }) + .AddToPlan(plan.get())); + state.ResumeTiming(); + ASSERT_FINISHES_OK(StartAndCollect(plan.get(), sink_gen)); + } + state.counters["rows_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches * batch_size), + benchmark::Counter::kIsRate); + + state.counters["batches_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches), benchmark::Counter::kIsRate); +} + +static void FilterOverheadIsolated(benchmark::State& state, Expression expr) { + const int32_t batch_size = static_cast<int32_t>(state.range(0)); + const int32_t num_batches = kTotalBatchSize / batch_size; + + arrow::compute::BatchesWithSchema data = MakeRandomBatches( + schema({field("i64", int64()), field("bool", boolean())}), num_batches, batch_size); + ExecContext ctx(default_memory_pool(), arrow::internal::GetCpuThreadPool()); + for (auto _ : state) { + state.PauseTiming(); + AsyncGenerator<util::optional<ExecBatch>> sink_gen; + + ASSERT_OK_AND_ASSIGN(std::shared_ptr<arrow::compute::ExecPlan> plan, + ExecPlan::Make(&ctx)); + // Source and sink nodes have no effect on the benchmark. + // Used for dummy purposes as they are referenced in InputReceived and InputFinished. + ASSERT_OK_AND_ASSIGN( + arrow::compute::ExecNode * source_node, + MakeExecNode("source", plan.get(), {}, + SourceNodeOptions{data.schema, data.gen(/*parallel=*/true, + /*slow=*/false)})); + ASSERT_OK_AND_ASSIGN(arrow::compute::ExecNode * project_node, + MakeExecNode("filter", plan.get(), {source_node}, + FilterNodeOptions{ + expr, + })); + MakeExecNode("sink", plan.get(), {project_node}, SinkNodeOptions{&sink_gen}); + + std::unique_ptr<arrow::compute::TaskScheduler> scheduler = TaskScheduler::Make(); + std::condition_variable all_tasks_finished_cv; + std::mutex mutex; + int task_group_id = scheduler->RegisterTaskGroup( + [&](size_t thread_id, int64_t task_id) { + project_node->InputReceived(source_node, data.batches[task_id]); + return Status::OK(); + }, + [&](size_t thread_id) { + project_node->InputFinished(source_node, static_cast<int>(data.batches.size())); + std::unique_lock<std::mutex> lk(mutex); + all_tasks_finished_cv.notify_one(); + return Status::OK(); + }); + scheduler->RegisterEnd(); + ThreadIndexer thread_indexer; + + state.ResumeTiming(); + arrow::internal::ThreadPool* thread_pool = arrow::internal::GetCpuThreadPool(); + ASSERT_OK(scheduler->StartScheduling( + thread_indexer(), + [&](std::function<Status(size_t)> task) -> Status { + return thread_pool->Spawn([&, task]() { + size_t tid = thread_indexer(); + ARROW_DCHECK_OK(task(tid)); + }); + }, + thread_pool->GetCapacity(), + /*use_sync_execution=*/false)); + std::unique_lock<std::mutex> lk(mutex); + ASSERT_OK(scheduler->StartTaskGroup(thread_indexer(), task_group_id, num_batches)); + all_tasks_finished_cv.wait(lk); + ASSERT_TRUE(project_node->finished().is_finished()); + } + state.counters["rows_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches * batch_size), + benchmark::Counter::kIsRate); + + state.counters["batches_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches), benchmark::Counter::kIsRate); +} + +arrow::compute::Expression complex_expression = + less(less(field_ref("i64"), literal(20)), greater(field_ref("i64"), literal(0))); +arrow::compute::Expression simple_expression = + less(call("negate", {field_ref("i64")}), literal(0)); +arrow::compute::Expression zero_copy_expression = is_valid((call( + "cast", {field_ref("i64")}, compute::CastOptions::Safe(timestamp(TimeUnit::NANO))))); Review Comment: I don't think we need zero copy for the filter benchmark. We can probably just get rid of this one. ########## cpp/src/arrow/compute/exec/filter_benchmark.cc: ########## @@ -0,0 +1,176 @@ +// 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 <condition_variable> +#include <mutex> + +#include "benchmark/benchmark.h" + +#include "arrow/compute/cast.h" +#include "arrow/compute/exec.h" +#include "arrow/compute/exec/expression.h" +#include "arrow/compute/exec/options.h" +#include "arrow/compute/exec/task_util.h" +#include "arrow/compute/exec/test_util.h" +#include "arrow/dataset/partition.h" +#include "arrow/testing/future_util.h" +#include "arrow/testing/generator.h" +#include "arrow/testing/gtest_util.h" +#include "arrow/type.h" + +namespace arrow { +namespace compute { + +static constexpr int64_t kTotalBatchSize = 1000000; + +static void FilterOverhead(benchmark::State& state, Expression expr) { + const int32_t batch_size = static_cast<int32_t>(state.range(0)); + const int32_t num_batches = kTotalBatchSize / batch_size; + + arrow::compute::BatchesWithSchema data = MakeRandomBatches( + schema({field("i64", int64()), field("bool", boolean())}), num_batches, batch_size); + ExecContext ctx(default_memory_pool(), arrow::internal::GetCpuThreadPool()); + for (auto _ : state) { + state.PauseTiming(); + ASSERT_OK_AND_ASSIGN(std::shared_ptr<arrow::compute::ExecPlan> plan, + ExecPlan::Make(&ctx)); + AsyncGenerator<util::optional<ExecBatch>> sink_gen; + ASSERT_OK(Declaration::Sequence( + { + {"source", + SourceNodeOptions{data.schema, + data.gen(/*parallel=*/true, /*slow=*/false)}, + "custom_source_label"}, + {"filter", + FilterNodeOptions{ + expr, + }}, + {"sink", SinkNodeOptions{&sink_gen}, "custom_sink_label"}, + }) + .AddToPlan(plan.get())); + state.ResumeTiming(); + ASSERT_FINISHES_OK(StartAndCollect(plan.get(), sink_gen)); + } + state.counters["rows_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches * batch_size), + benchmark::Counter::kIsRate); + + state.counters["batches_per_second"] = benchmark::Counter( + static_cast<double>(state.iterations() * num_batches), benchmark::Counter::kIsRate); +} + +static void FilterOverheadIsolated(benchmark::State& state, Expression expr) { + const int32_t batch_size = static_cast<int32_t>(state.range(0)); + const int32_t num_batches = kTotalBatchSize / batch_size; + + arrow::compute::BatchesWithSchema data = MakeRandomBatches( + schema({field("i64", int64()), field("bool", boolean())}), num_batches, batch_size); + ExecContext ctx(default_memory_pool(), arrow::internal::GetCpuThreadPool()); + for (auto _ : state) { + state.PauseTiming(); + AsyncGenerator<util::optional<ExecBatch>> sink_gen; + + ASSERT_OK_AND_ASSIGN(std::shared_ptr<arrow::compute::ExecPlan> plan, + ExecPlan::Make(&ctx)); + // Source and sink nodes have no effect on the benchmark. + // Used for dummy purposes as they are referenced in InputReceived and InputFinished. + ASSERT_OK_AND_ASSIGN( + arrow::compute::ExecNode * source_node, + MakeExecNode("source", plan.get(), {}, + SourceNodeOptions{data.schema, data.gen(/*parallel=*/true, + /*slow=*/false)})); + ASSERT_OK_AND_ASSIGN(arrow::compute::ExecNode * project_node, Review Comment: ```suggestion ASSERT_OK_AND_ASSIGN(arrow::compute::ExecNode * filter_node, ``` Alternatively, we might want to see if we can create a benchmark_util.cc/h that we can use to share some of this logic that will be identical between different benchmarks. Though maybe filter & project are the only ones simple enough to share the same basic structure. Also, long term, I think the `XyzIsolated` benchmarks will be less useful, as we address the scheduling issues, so maybe we don't need to spend too much time optimizing the case. Still, it would be kind of nice to have less code that feels copy/pasted between the benchmarks. -- 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: github-unsubscr...@arrow.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
