sunchao commented on code in PR #100:
URL:
https://github.com/apache/arrow-datafusion-comet/pull/100#discussion_r1503764067
##########
spark/src/main/scala/org/apache/comet/CometSparkSessionExtensions.scala:
##########
@@ -451,6 +468,23 @@ class CometSparkSessionExtensions
}
}
}
+
+ // CometExec already wraps a `ColumnarToRowExec` for row-based operators.
Therefore,
+ // `ColumnarToRowExec` is redundant and can be eliminated.
+ //
+ // It was added during ApplyColumnarRulesAndInsertTransitions'
insertTransitions phase when Spark
+ // requests row-based output such as `collect` call. It's correct to add a
redundant
+ // `ColumnarToRowExec` for `CometExec`. However, for certain operators such
as
+ // `CometCollectLimitExec` which overrides `executeCollect`, the redundant
`ColumnarToRowExec`
+ // makes the override ineffective. The purpose of this rule is to eliminate
the redundant
+ // `ColumnarToRowExec` for such operators.
+ case class EliminateRedundantColumnarToRow(session: SparkSession) extends
Rule[SparkPlan] {
Review Comment:
Hmm I'm trying to understand why this is necessary. The test passes even if
I remove this rule.
##########
spark/src/main/scala/org/apache/spark/sql/comet/CometCollectLimitExec.scala:
##########
@@ -0,0 +1,113 @@
+/**
+ * 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.
+ */
+package org.apache.spark.sql.comet
+
+import java.util.Objects
+
+import org.apache.spark.rdd.RDD
+import org.apache.spark.serializer.Serializer
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.comet.execution.shuffle.{CometShuffledBatchRDD,
CometShuffleExchangeExec}
+import org.apache.spark.sql.execution.{ColumnarToRowExec, SparkPlan,
UnaryExecNode, UnsafeRowSerializer}
+import org.apache.spark.sql.execution.metric.{SQLMetric, SQLMetrics,
SQLShuffleReadMetricsReporter, SQLShuffleWriteMetricsReporter}
+import org.apache.spark.sql.vectorized.ColumnarBatch
+
+/**
+ * Comet physical plan node for Spark `CollectExecNode`.
Review Comment:
nit: `CollectExecNode` -> `CollectLimitExec`?
##########
spark/src/main/scala/org/apache/spark/sql/comet/CometCollectLimitExec.scala:
##########
@@ -0,0 +1,113 @@
+/**
+ * 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.
+ */
+package org.apache.spark.sql.comet
+
+import java.util.Objects
+
+import org.apache.spark.rdd.RDD
+import org.apache.spark.serializer.Serializer
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.comet.execution.shuffle.{CometShuffledBatchRDD,
CometShuffleExchangeExec}
+import org.apache.spark.sql.execution.{ColumnarToRowExec, SparkPlan,
UnaryExecNode, UnsafeRowSerializer}
+import org.apache.spark.sql.execution.metric.{SQLMetric, SQLMetrics,
SQLShuffleReadMetricsReporter, SQLShuffleWriteMetricsReporter}
+import org.apache.spark.sql.vectorized.ColumnarBatch
+
+/**
+ * Comet physical plan node for Spark `CollectExecNode`.
+ *
+ * Similar to `CometTakeOrderedAndProjectExec`, it contains two native
executions seperated by a
+ * comet shuffle.
+ */
+case class CometCollectLimitExec(
+ override val originalPlan: SparkPlan,
+ limit: Int,
+ offset: Int,
+ child: SparkPlan)
+ extends CometExec
+ with UnaryExecNode {
+
+ private lazy val writeMetrics =
+ SQLShuffleWriteMetricsReporter.createShuffleWriteMetrics(sparkContext)
+ private lazy val readMetrics =
+ SQLShuffleReadMetricsReporter.createShuffleReadMetrics(sparkContext)
+ override lazy val metrics: Map[String, SQLMetric] = Map(
+ "dataSize" -> SQLMetrics.createSizeMetric(sparkContext, "data size"),
+ "shuffleReadElapsedCompute" ->
+ SQLMetrics.createNanoTimingMetric(sparkContext, "shuffle read elapsed
compute at native"),
+ "numPartitions" -> SQLMetrics.createMetric(
+ sparkContext,
+ "number of partitions")) ++ readMetrics ++ writeMetrics
+
+ private lazy val serializer: Serializer =
+ new UnsafeRowSerializer(child.output.size, longMetric("dataSize"))
+
+ override def executeCollect(): Array[InternalRow] = {
+ ColumnarToRowExec(child).executeTake(limit)
+ }
+
+ protected override def doExecuteColumnar(): RDD[ColumnarBatch] = {
+ val childRDD = child.executeColumnar()
+ if (childRDD.getNumPartitions == 0) {
+ CometExecUtils.createEmptyColumnarRDDWithSinglePartition(sparkContext)
+ } else {
+ val singlePartitionRDD = if (childRDD.getNumPartitions == 1) {
+ childRDD
+ } else {
+ val localLimitedRDD = if (limit >= 0) {
+ childRDD.mapPartitionsInternal { iter =>
+ val limitOp = CometExecUtils.getLimitNativePlan(output, limit).get
+ CometExec.getCometIterator(Seq(iter), limitOp)
+ }
+ } else {
+ childRDD
+ }
+ // Shuffle to Single Partition using Comet native shuffle
+ val dep = CometShuffleExchangeExec.prepareShuffleDependency(
+ localLimitedRDD,
+ child.output,
+ outputPartitioning,
+ serializer,
+ metrics)
+ metrics("numPartitions").set(dep.partitioner.numPartitions)
+
+ new CometShuffledBatchRDD(dep, readMetrics)
+ }
+
+ // todo: supports offset later
Review Comment:
nit: `todo` -> `TODO`
Looks like
[`GlobalLimitExec`](https://github.com/apache/arrow-datafusion/blob/main/datafusion/physical-plan/src/limit.rs#L47)
in DF already supports offset, and maybe we can use it later. Currently we are
using `LocalLimitExec` on the native side.
##########
spark/src/main/scala/org/apache/spark/sql/comet/CometCollectLimitExec.scala:
##########
@@ -0,0 +1,113 @@
+/**
+ * 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.
+ */
+package org.apache.spark.sql.comet
+
+import java.util.Objects
+
+import org.apache.spark.rdd.RDD
+import org.apache.spark.serializer.Serializer
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.comet.execution.shuffle.{CometShuffledBatchRDD,
CometShuffleExchangeExec}
+import org.apache.spark.sql.execution.{ColumnarToRowExec, SparkPlan,
UnaryExecNode, UnsafeRowSerializer}
+import org.apache.spark.sql.execution.metric.{SQLMetric, SQLMetrics,
SQLShuffleReadMetricsReporter, SQLShuffleWriteMetricsReporter}
+import org.apache.spark.sql.vectorized.ColumnarBatch
+
+/**
+ * Comet physical plan node for Spark `CollectExecNode`.
+ *
+ * Similar to `CometTakeOrderedAndProjectExec`, it contains two native
executions seperated by a
+ * comet shuffle.
Review Comment:
nit: `comet` -> `Comet`
##########
spark/src/main/scala/org/apache/comet/shims/ShimCometSparkSessionExtensions.scala:
##########
@@ -32,4 +33,9 @@ trait ShimCometSparkSessionExtensions {
.map { a => a.setAccessible(true); a }
.flatMap(_.get(scan).asInstanceOf[Option[Aggregation]])
.headOption
+
+ def getOffset(limit: LimitExec): Option[Int] =
limit.getClass.getDeclaredFields
Review Comment:
nit: I wonder if we can just return `0` if there is no offset in this
method, so that we don't have to do `getOffset(op).getOrElse(0)` in a few
places.
##########
spark/src/main/scala/org/apache/spark/sql/comet/CometCollectLimitExec.scala:
##########
@@ -0,0 +1,113 @@
+/**
+ * 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.
+ */
Review Comment:
nit: I think we usually keep a blank line between the header and the
`package`.
--
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]
