Count expressions per Exec in SQLPlanParser

Signed-off-by: Ahmed Hussein (amahussein) <[email protected]> Fixes NVIDIA#1447 Improves the operators-stats by counting the occurence of each expression within the Exec node.
amahussein · Dec 5, 2024 · c15aac9 · c15aac9
1 parent a414e09
commit c15aac9
Show file tree

Hide file tree

Showing 10 changed files with 242 additions and 110 deletions.
diff --git a/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/SQLPlanParser.scala b/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/SQLPlanParser.scala
@@ -21,7 +21,7 @@ import scala.collection.mutable.{ArrayBuffer, WeakHashMap}
 import scala.util.control.NonFatal
 import scala.util.matching.Regex
 
-import com.nvidia.spark.rapids.tool.planparser.ops.{OperatorRefBase, OpRef, UnsupportedExprOpRef}
+import com.nvidia.spark.rapids.tool.planparser.ops.{ExprOpRef, OperatorRefTrait, OpRef, UnsupportedExprOpRef}
 import com.nvidia.spark.rapids.tool.planparser.photon.{PhotonPlanParser, PhotonStageExecParser}
 import com.nvidia.spark.rapids.tool.qualification.PluginTypeChecker
 
@@ -75,7 +75,7 @@ object UnsupportedReasons extends Enumeration {
 case class UnsupportedExecSummary(
     sqlId: Long,
     execId: Long,
-    execRef: OperatorRefBase,
+    execRef: OperatorRefTrait,
     opType: OpTypes.OpType,
     reason: UnsupportedReasons.UnsupportedReason,
     opAction: OpActions.OpAction) {
@@ -89,8 +89,6 @@ case class UnsupportedExecSummary(
   val unsupportedOperatorCSVFormat: String = execRef.getOpNameCSV
 
   val details: String = UnsupportedReasons.reportUnsupportedReason(reason)
-
-  def isExpression: Boolean = execRef.isInstanceOf[UnsupportedExprOpRef]
 }
 
 case class ExecInfo(
@@ -110,7 +108,7 @@ case class ExecInfo(
     dataSet: Boolean,
     udf: Boolean,
     shouldIgnore: Boolean,
-    expressions: Seq[OpRef]) {
+    expressions: Seq[ExprOpRef]) {
 
   private def childrenToString = {
     val str = children.map { c =>
@@ -141,10 +139,6 @@ case class ExecInfo(
     stages = stageIDs
   }
 
-  def appendToStages(stageIDs: Set[Int]): Unit = {
-    stages ++= stageIDs
-  }
-
   def setShouldRemove(value: Boolean): Unit = {
     shouldRemove ||= value
   }
@@ -286,7 +280,7 @@ object ExecInfo {
       udf,
       shouldIgnore,
       // convert array of string expressions to OpRefs
-      expressions = expressions.map(OpRef.fromExpr)
+      expressions = ExprOpRef.fromRawExprSeq(expressions)
     )
   }
 
@@ -351,7 +345,7 @@ case class PlanInfo(
     sqlID: Long,
     sqlDesc: String,
     execInfo: Seq[ExecInfo]) {
-  def getUnsupportedExpressions: Seq[OperatorRefBase] = {
+  def getUnsupportedExpressions: Seq[OperatorRefTrait] = {
     execInfo.flatMap { e =>
       if (e.isClusterNode) {
         // wholeStageCodeGen does not have expressions/unsupported-expressions
@@ -727,21 +721,21 @@ object SQLPlanParser extends Logging {
   }
 
   private def getAllFunctionNames(regPattern: Regex, expr: String,
-      groupInd: Int = 1, isAggr: Boolean = true): Set[String] = {
+      groupInd: Int = 1, isAggr: Boolean = true): Array[String] = {
     // Returns all matches in an expression. This can be used when the SQL expression is not
     // tokenized.
     val newExpr = processSpecialFunctions(expr)
 
     // first get all the functionNames
     val exprss =
-      regPattern.findAllMatchIn(newExpr).map(_.group(groupInd)).toSet
+      regPattern.findAllMatchIn(newExpr).map(_.group(groupInd)).toSeq
 
     // For aggregate expressions we want to process the results to remove the prefix
     // DB: remove the "^partial_" and "^finalmerge_" prefixes
     // TODO:
     //    for performance sake, we can turn off the aggregate processing by enabling it only
     //    when needed. However, for now, we always do this processing until we are confident we know
-    //    the correct place to turn on/off that flag.we can use the argument isAgg only when needed
+    //    the correct place to turn on/off that flag. We can use the argument isAgg only when needed
     val results = if (isAggr) {
       exprss.collect {
         case func =>
@@ -750,15 +744,15 @@ object SQLPlanParser extends Logging {
     } else {
       exprss
     }
-    results.filterNot(ignoreExpression(_))
+    results.filterNot(ignoreExpression(_)).toArray
   }
 
   def parseProjectExpressions(exprStr: String): Array[String] = {
     // Project [cast(value#136 as string) AS value#144, CEIL(value#136) AS CEIL(value)#143L]
     // This is to split the string such that only function names are extracted. The pattern is
     // such that function name is succeeded by `(`. We use regex to extract all the function names
     // below:
-    getAllFunctionNames(functionPrefixPattern, exprStr).toArray
+    getAllFunctionNames(functionPrefixPattern, exprStr)
   }
 
   // This parser is used for SortAggregateExec, HashAggregateExec and ObjectHashAggregateExec
@@ -792,7 +786,7 @@ object SQLPlanParser extends Logging {
         }
       }
     }
-    parsedExpressions.distinct.toArray
+    parsedExpressions.toArray
   }
 
   def parseWindowExpressions(exprStr:String): Array[String] = {
@@ -828,7 +822,7 @@ object SQLPlanParser extends Logging {
         }
       }
     }
-    parsedExpressions.distinct.toArray
+    parsedExpressions.toArray
   }
 
   def parseWindowGroupLimitExpressions(exprStr: String): Array[String] = {
@@ -858,10 +852,10 @@ object SQLPlanParser extends Logging {
     //  - Some values can be NULLs. That's why we cannot limit the extract to the first row.
     //  - Nested brackets/parenthesis makes it challenging to use regex that contains
     //    brackets/parenthesis to extract expressions.
-    // The implementation Use regex to extract all function names and return distinct set of
+    // The implementation Use regex to extract all function names and return a list of
     // function names.
     // This implementation is 1 line implementation, but it can be a memory/time bottleneck.
-    getAllFunctionNames(functionPrefixPattern, exprStr).toArray
+    getAllFunctionNames(functionPrefixPattern, exprStr)
   }
 
   def parseTakeOrderedExpressions(exprStr: String): Array[String] = {
@@ -889,19 +883,19 @@ object SQLPlanParser extends Logging {
         }
       }
     }
-    parsedExpressions.distinct.toArray
+    parsedExpressions.toArray
   }
 
   def parseGenerateExpressions(exprStr: String): Array[String] = {
     // Get the function names from the GenerateExec. The GenerateExec has the following format:
     // 1. Generate explode(arrays#1306), [id#1304], true, [col#1426]
     // 2. Generate json_tuple(values#1305, Zipcode, ZipCodeType, City), [id#1304],
     // false, [c0#1407, c1#1408, c2#1409]
-    getAllFunctionNames(functionPrefixPattern, exprStr).toArray
+    getAllFunctionNames(functionPrefixPattern, exprStr)
   }
 
   private def addFunctionNames(exprs: String, parsedExpressions: ArrayBuffer[String]): Unit = {
-    val functionNames = getAllFunctionNames(functionPrefixPattern, exprs).toArray
+    val functionNames = getAllFunctionNames(functionPrefixPattern, exprs)
     functionNames.foreach(parsedExpressions += _)
   }
 
@@ -946,7 +940,7 @@ object SQLPlanParser extends Logging {
     val isSortMergeSupported = !(isSortMergeJoin &&
         joinCondition.nonEmpty && isSMJConditionUnsupported(joinCondition))
 
-    (parsedExpressions.distinct.toArray, equiJoinSupportedTypes(buildSide, joinType)
+    (parsedExpressions.toArray, equiJoinSupportedTypes(buildSide, joinType)
         && isSortMergeSupported)
   }
 
@@ -1054,7 +1048,7 @@ object SQLPlanParser extends Logging {
         }
       }
     }
-    parsedExpressions.distinct.toArray
+    parsedExpressions.toArray
   }
 
   def parseFilterExpressions(exprStr: String): Array[String] = {
@@ -1109,7 +1103,7 @@ object SQLPlanParser extends Logging {
     processedExpr = nonBinaryOperatorsRegEx.replaceAllIn(processedExpr, " ")
 
     // Step-4: remove remaining parentheses '(', ')' and commas if we had functionCalls
-    if (!functionMatches.isEmpty) {
+    if (functionMatches.nonEmpty) {
       // remove ","
       processedExpr = processedExpr.replaceAll(",", " ")
     }
@@ -1147,7 +1141,6 @@ object SQLPlanParser extends Logging {
           logDebug(s"Unrecognized Token - $token")
       }
     }
-
-    parsedExpressions.distinct.toArray
+    parsedExpressions.toArray
   }
 }
diff --git a/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/WholeStageExecParser.scala b/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/WholeStageExecParser.scala
@@ -59,7 +59,7 @@ abstract class WholeStageExecParserBase(
     // The node should be marked as shouldRemove when all the children of the
     // wholeStageCodeGen are marked as shouldRemove.
     val removeNode = isDupNode || childNodes.forall(_.shouldRemove)
-    // Remove any suffix in order to get the node label without any trailing number.
+    // Remove any suffix to get the node label without any trailing number.
     val nodeLabel = nodeNameRegeX.findFirstMatchIn(node.name) match {
       case Some(m) => m.group(1)
       // in case not found, use the full exec name

diff --git a/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/WindowGroupLimitParser.scala b/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/WindowGroupLimitParser.scala
@@ -37,7 +37,7 @@ case class WindowGroupLimitParser(
 
   override def getUnsupportedExprReasonsForExec(
       expressions: Array[String]): Seq[UnsupportedExprOpRef] = {
-    expressions.flatMap { expr =>
+    expressions.distinct.flatMap { expr =>
       if (!supportedRankingExprs.contains(expr)) {
         Some(UnsupportedExprOpRef(expr,
           s"Ranking function $expr is not supported in $fullExecName"))

diff --git a/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/ops/ExprOpRef.scala b/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/ops/ExprOpRef.scala
@@ -0,0 +1,31 @@
+/*
+ * Copyright (c) 2024, NVIDIA CORPORATION.
+ *
+ * Licensed 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 com.nvidia.spark.rapids.tool.planparser.ops
+
+/**
+ * Represents a reference to an expression operator that is stored in the ExecInfo expressions
+ * @param opRef the opRef to wrap
+ * @param count the count of that expression within the exec.
+ */
+case class ExprOpRef(opRef: OpRef, count: Int = 1) extends OpRefWrapperBase(opRef)
+
+object ExprOpRef extends OpRefWrapperBaseTrait[ExprOpRef] {
+  def fromRawExprSeq(exprArr: Seq[String]): Seq[ExprOpRef] = {
+    exprArr.groupBy(identity)
+      .mapValues(expr => ExprOpRef(OpRef.fromExpr(expr.head), expr.size)).values.toSeq
+  }
+}
diff --git a/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/ops/OpRefWrapperBase.scala b/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/ops/OpRefWrapperBase.scala
@@ -0,0 +1,50 @@
+/*
+ * Copyright (c) 2024, NVIDIA CORPORATION.
+ *
+ * Licensed 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 com.nvidia.spark.rapids.tool.planparser.ops
+
+/**
+ * An instance that wraps OpRef and exposes the OpRef methods.
+ * This is used to provide common interface for all classes that wrap OpRef along with other
+ * metadata.
+ * @param opRef the opRef to wrap
+ */
+class OpRefWrapperBase(opRef: OpRef) extends OperatorRefTrait {
+  override def getOpName: String = opRef.getOpName
+
+  override def getOpNameCSV: String = opRef.getOpNameCSV
+
+  override def getOpType: String = opRef.getOpType
+
+  override def getOpTypeCSV: String = opRef.getOpTypeCSV
+
+  def getOpRef: OpRef = opRef
+}
+
+/**
+ * A trait that provides a factory method to create instances of OpRefWrapperBase from a sequence of
+ * @tparam R the type of the OpRefWrapperBase
+ */
+trait OpRefWrapperBaseTrait[R <: OpRefWrapperBase] {
+  /**
+   * Create instances of OpRefWrapperBase from a sequence of expressions.
+   * The expressions are grouped by their value and the count of each expression is stored in the
+   * OpRefWrapperBase entry.
+   * @param exprArr the sequence of expressions
+   * @return a sequence of OpRefWrapperBase instances
+   */
+  def fromRawExprSeq(exprArr: Seq[String]): Seq[R]
+}
diff --git a/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/ops/OperatorCounter.scala b/core/src/main/scala/com/nvidia/spark/rapids/tool/planparser/ops/OperatorCounter.scala
@@ -38,66 +38,71 @@ case class OperatorCounter(planInfo: PlanInfo) {
    * @param stages The set of stages where the operator appears.
    */
   case class OperatorData(
-      opRef: OperatorRefBase,
+      opRef: OpRef,
       var count: Int = 0,
-      var stages: Set[Int] = Set())
+      var stages: Set[Int] = Set()) extends OpRefWrapperBase(opRef)
 
   // Summarizes the count information for an exec or expression, including whether it is supported.
   case class OperatorCountSummary(
     opData: OperatorData,
     isSupported: Boolean)
 
-  private val supportedMap: mutable.Map[OperatorRefBase, OperatorData] = mutable.Map()
-  private val unsupportedMap: mutable.Map[OperatorRefBase, OperatorData] = mutable.Map()
+  private val supportedMap: mutable.Map[OpRef, OperatorData] = mutable.Map()
+  private val unsupportedMap: mutable.Map[OpRef, OperatorData] = mutable.Map()
 
   // Returns a sequence of `OperatorCountSummary`, combining both supported and
   // unsupported operators.
-  def getOpsCountSummary(): Seq[OperatorCountSummary] = {
+  def getOpsCountSummary: Seq[OperatorCountSummary] = {
     supportedMap.values.map(OperatorCountSummary(_, isSupported = true)).toSeq ++
       unsupportedMap.values.map(OperatorCountSummary(_, isSupported = false)).toSeq
   }
 
 
   // Updates the operator data in the given map (supported or unsupported).
   // Increments the count and updates the stages where the operator appears.
-  private def updateOpRefEntry(opRef: OperatorRefBase, stages: Set[Int],
-    targetMap: mutable.Map[OperatorRefBase, OperatorData]): Unit = {
+  private def updateOpRefEntry(opRef: OpRef, stages: Set[Int],
+    targetMap: mutable.Map[OpRef, OperatorData], incrValue: Int = 1): Unit = {
     val operatorData = targetMap.getOrElseUpdate(opRef, OperatorData(opRef))
-    operatorData.count += 1
+    operatorData.count += incrValue
     operatorData.stages ++= stages
   }
 
   // Processes an `ExecInfo` node to update exec and expression counts.
   // Separates supported and unsupported execs and expressions into their respective maps.
   private def processExecInfo(execInfo: ExecInfo): Unit = {
-    val opMap = execInfo.isSupported match {
-      case true => supportedMap
-      case false => unsupportedMap
+    val opMap = if (execInfo.isSupported) {
+      supportedMap
+    } else {
+      unsupportedMap
     }
     updateOpRefEntry(execInfo.execRef, execInfo.stages, opMap)
-    // update the map for supported expressions. We should exclude the unsupported expressions.
-    execInfo.expressions.filterNot(
-      e => execInfo.unsupportedExprs.exists(exp => exp.opRef.equals(e))).foreach { expr =>
-      updateOpRefEntry(expr, execInfo.stages, supportedMap)
-    }
-    // update the map for unsupported expressions
-    execInfo.unsupportedExprs.foreach { expr =>
-      updateOpRefEntry(expr, execInfo.stages, unsupportedMap)
+    // Update the map for supported expressions. For unsupported expressions,
+    // we use the count stored in the supported expressions.
+    execInfo.expressions.foreach { expr =>
+      val exprMap =
+        if (execInfo.unsupportedExprs.exists(unsupExec =>
+          unsupExec.getOpRef.equals(expr.getOpRef))) {
+          // The expression skips because it exists in the unsupported expressions.
+          unsupportedMap
+        } else {
+          supportedMap
+        }
+      updateOpRefEntry(expr.getOpRef, execInfo.stages, exprMap, expr.count)
     }
   }
 
-  // Counts the execs and expressions in the execution plan.
+  // Counts the execs and expressions in the execution plan excluding clusterNodes
+  // (i.e., WholeStageCodeGen).
   private def countOperators(): Unit = {
     planInfo.execInfo.foreach { exec =>
-      exec.isClusterNode match {
-        // we do not want to count the cluster nodes in that aggregation
-        case true =>
-          if (exec.children.nonEmpty) {
-            exec.children.get.foreach { child =>
-              processExecInfo(child)
-            }
+      if (exec.isClusterNode) {
+        if (exec.children.nonEmpty) {
+          exec.children.get.foreach { child =>
+            processExecInfo(child)
           }
-        case false => processExecInfo(exec)
+        }
+      } else {
+        processExecInfo(exec)
       }
     }
   }