Copilot commented on code in PR #13357: URL: https://github.com/apache/gluten/pull/13357#discussion_r4360646469
########## cpp/velox/operators/functions/IncrementMetricFunction.cc: ########## @@ -0,0 +1,114 @@ +/* + * 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 "operators/functions/IncrementMetricFunction.h" + +#include <memory> +#include <vector> + +#include "velox/core/Expressions.h" +#include "velox/core/QueryConfig.h" +#include "velox/expression/EvalCtx.h" +#include "velox/expression/FunctionSignature.h" +#include "velox/expression/VectorFunction.h" + +namespace gluten { +namespace { + +using namespace facebook::velox; + +/// Returns its only argument. See the header for why the name, not the body, carries the meaning. +class IncrementMetricFunction : public exec::VectorFunction { + public: + void apply( + const SelectivityVector& rows, + std::vector<VectorPtr>& args, + const TypePtr& /*outputType*/, + exec::EvalCtx& context, + VectorPtr& result) const override { + VELOX_CHECK_EQ(args.size(), 1); + context.moveOrCopyResult(args[0], rows, result); + } +}; + +bool callsIncrementMetric(const core::ITypedExpr& expr) { + if (const auto* call = dynamic_cast<const core::CallTypedExpr*>(&expr)) { + if (isIncrementMetricFunction(call->name())) { + return true; + } + } + if (const auto* lambda = dynamic_cast<const core::LambdaTypedExpr*>(&expr)) { + return lambda->body() != nullptr && callsIncrementMetric(*lambda->body()); + } + for (const auto& input : expr.inputs()) { + if (input != nullptr && callsIncrementMetric(*input)) { + return true; + } + } + return false; +} + +} // namespace + +std::string incrementMetricFunctionName(int32_t slot) { + return std::string(kIncrementMetricFunctionPrefix) + std::to_string(slot); +} + +bool isIncrementMetricFunction(const std::string& functionName) { + return functionName.rfind(kIncrementMetricFunctionPrefix, 0) == 0; +} Review Comment: `isIncrementMetricFunction` currently treats any name with the `increment_metric_` prefix as a counter (e.g., `increment_metric_foo`, `increment_metric_999`, or even `increment_metric_`). This can incorrectly (a) force query configs for plans that aren’t using the fixed slots and (b) export unrelated expression stats under `expressionStats`. Consider validating that the suffix is an integer and within `[0, kIncrementMetricFunctionSlots)` (and optionally rejecting negative / missing suffix). ########## cpp/velox/operators/functions/IncrementMetricFunction.cc: ########## @@ -0,0 +1,114 @@ +/* + * 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 "operators/functions/IncrementMetricFunction.h" + +#include <memory> +#include <vector> + +#include "velox/core/Expressions.h" +#include "velox/core/QueryConfig.h" +#include "velox/expression/EvalCtx.h" +#include "velox/expression/FunctionSignature.h" +#include "velox/expression/VectorFunction.h" + +namespace gluten { +namespace { + +using namespace facebook::velox; + +/// Returns its only argument. See the header for why the name, not the body, carries the meaning. +class IncrementMetricFunction : public exec::VectorFunction { + public: + void apply( + const SelectivityVector& rows, + std::vector<VectorPtr>& args, + const TypePtr& /*outputType*/, + exec::EvalCtx& context, + VectorPtr& result) const override { + VELOX_CHECK_EQ(args.size(), 1); + context.moveOrCopyResult(args[0], rows, result); + } +}; + +bool callsIncrementMetric(const core::ITypedExpr& expr) { + if (const auto* call = dynamic_cast<const core::CallTypedExpr*>(&expr)) { + if (isIncrementMetricFunction(call->name())) { + return true; + } + } + if (const auto* lambda = dynamic_cast<const core::LambdaTypedExpr*>(&expr)) { + return lambda->body() != nullptr && callsIncrementMetric(*lambda->body()); + } + for (const auto& input : expr.inputs()) { + if (input != nullptr && callsIncrementMetric(*input)) { + return true; + } + } + return false; +} + +} // namespace + +std::string incrementMetricFunctionName(int32_t slot) { + return std::string(kIncrementMetricFunctionPrefix) + std::to_string(slot); +} + +bool isIncrementMetricFunction(const std::string& functionName) { + return functionName.rfind(kIncrementMetricFunctionPrefix, 0) == 0; +} + +void registerIncrementMetricFunctions() { + // Not deterministic: keeps Velox from constant-folding or dictionary-peeling the call, either of + // which would change how many rows it is evaluated on. No default null behavior: a null input + // still counts, as it does on Spark. + const auto metadata = exec::VectorFunctionMetadataBuilder().deterministic(false).defaultNullBehavior(false).build(); + for (int32_t slot = 0; slot < kIncrementMetricFunctionSlots; ++slot) { + exec::registerVectorFunction( + incrementMetricFunctionName(slot), + {exec::FunctionSignatureBuilder().typeVariable("T").returnType("T").argumentType("T").build()}, + std::make_unique<IncrementMetricFunction>(), + metadata); + } +} + +bool usesIncrementMetricFunctions(const core::PlanNode& plan) { + if (const auto* project = dynamic_cast<const core::ProjectNode*>(&plan)) { + for (const auto& projection : project->projections()) { + if (projection != nullptr && callsIncrementMetric(*projection)) { + return true; + } + } + } else if (const auto* filter = dynamic_cast<const core::FilterNode*>(&plan)) { + if (filter->filter() != nullptr && callsIncrementMetric(*filter->filter())) { + return true; + } + } + for (const auto& source : plan.sources()) { + if (source != nullptr && usesIncrementMetricFunctions(*source)) { + return true; + } + } + return false; +} Review Comment: Plan detection only inspects `ProjectNode` and `FilterNode`. If counters ever appear in other plan nodes that carry expressions (e.g., aggregations, joins, table writes, etc.), the required configs won’t be forced and counts can silently be wrong. If the intent is truly limited to project/filter, consider making that constraint explicit (e.g., fail fast when a counter call is found elsewhere). Otherwise, consider extending detection to inspect expressions for additional node types (or using a more generic plan/expr visitor where available). ########## cpp/velox/compute/WholeStageResultIterator.cc: ########## @@ -497,17 +499,26 @@ void WholeStageResultIterator::collectMetrics() { customStats[customMetric.first] = folly::dynamic::object("sum", customMetric.second.sum)( "count", customMetric.second.count)("min", customMetric.second.min)("max", customMetric.second.max); } + // Rows processed per Delta counter function, keyed by its slot name. Project and filter + // operators export these only when the plan calls a counter. + folly::dynamic expressionStats = folly::dynamic::object(); + for (const auto& exprStats : opStats->expressionStats) { + if (isIncrementMetricFunction(exprStats.first)) { + expressionStats[exprStats.first] = exprStats.second.numProcessedRows; + } + } - operatorStats.push_back(folly::dynamic::object("inputRows", opStats->inputRows)( - "inputVectors", opStats->inputVectors)("inputBytes", opStats->inputBytes)( - "rawInputRows", opStats->rawInputRows)("rawInputBytes", opStats->rawInputBytes)( - "outputRows", opStats->outputRows)("outputVectors", opStats->outputVectors)( - "outputBytes", opStats->outputBytes)("cpuCount", opStats->cpuWallTiming.count)( - "wallNanos", opStats->cpuWallTiming.wallNanos)("peakMemoryBytes", opStats->peakMemoryBytes)( - "numMemoryAllocations", opStats->numMemoryAllocations)("spilledInputBytes", opStats->spilledInputBytes)( - "spilledBytes", opStats->spilledBytes)("spilledRows", opStats->spilledRows)( - "spilledPartitions", opStats->spilledPartitions)("spilledFiles", opStats->spilledFiles)( - "physicalWrittenBytes", opStats->physicalWrittenBytes)("customStats", customStats)); + operatorStats.push_back( + folly::dynamic::object("inputRows", opStats->inputRows)("inputVectors", opStats->inputVectors)( + "inputBytes", opStats->inputBytes)("rawInputRows", opStats->rawInputRows)( + "rawInputBytes", opStats->rawInputBytes)("outputRows", opStats->outputRows)( + "outputVectors", opStats->outputVectors)("outputBytes", opStats->outputBytes)( + "cpuCount", opStats->cpuWallTiming.count)("wallNanos", opStats->cpuWallTiming.wallNanos)( + "peakMemoryBytes", opStats->peakMemoryBytes)("numMemoryAllocations", opStats->numMemoryAllocations)( + "spilledInputBytes", opStats->spilledInputBytes)("spilledBytes", opStats->spilledBytes)( + "spilledRows", opStats->spilledRows)("spilledPartitions", opStats->spilledPartitions)( + "spilledFiles", opStats->spilledFiles)("physicalWrittenBytes", opStats->physicalWrittenBytes)( + "customStats", customStats)("expressionStats", expressionStats)); Review Comment: This adds `expressionStats` to the operator metrics JSON unconditionally (often as an empty object). If any downstream consumer expects a strict schema, this may be a breaking change. If compatibility is a concern, consider only adding the `expressionStats` key when it’s non-empty (or when `trackExpressionStats_` is true), which also reduces payload size. ########## cpp/velox/compute/WholeStageResultIterator.h: ########## @@ -136,6 +136,9 @@ class WholeStageResultIterator : public SplitAwareColumnarBatchIterator { std::unique_ptr<facebook::velox::exec::TaskCursor> cursor_; facebook::velox::exec::Task* task_ = nullptr; std::shared_ptr<const facebook::velox::core::PlanNode> veloxPlan_; + /// Whether the plan calls a Delta counter and so needs the query configs set by + /// setIncrementMetricQueryConfigs(). + const bool trackExpressionStats_; Review Comment: `trackExpressionStats_` reads like a direct reflection of the query config, but the comment indicates it’s really ‘does this plan use IncrementMetric counters (and therefore require forcing config)?’. A name like `usesIncrementMetricFunctions_` or `needsIncrementMetricQueryConfigs_` would more precisely describe the stored value and reduce confusion. -- 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]
