[CARMEL-6375] Support Potential Skewed Operator Tagging (#1147)

apache · Nov 18, 2022 · c25fdd6 · c25fdd6
1 parent 3f2a8a9
commit c25fdd6
Show file tree

Hide file tree

Showing 2 changed files with 32 additions and 5 deletions.
diff --git a/sql/core/src/main/scala/org/apache/spark/sql/execution/DataSourceScanExec.scala b/sql/core/src/main/scala/org/apache/spark/sql/execution/DataSourceScanExec.scala
@@ -422,8 +422,9 @@ case class FileSourceScanExec(
         tableParallelInfo.get.partitionNumber.isEmpty))
   }
 
+  var _isBucketSkew: Option[Boolean] = None
   lazy val isBucketSkew = {
-    bucketedScan && {
+    _isBucketSkew = Some(bucketedScan && {
       val files = selectedPartitions.flatMap(partition => partition.files)
       val bucketFilesGroupingSize = files.map(file => (file.getPath.getName, file.getLen))
         .groupBy(file => BucketingUtils.getBucketId(file._1))
@@ -432,7 +433,8 @@ case class FileSourceScanExec(
       bucketFilesGroupingSize.nonEmpty &&
         SkewHandlingUtil.isSkewed(bucketFilesGroupingSize.max,
         Utils.median(bucketFilesGroupingSize, false), conf)
-    }
+    })
+    _isBucketSkew
   }
 
   override lazy val (outputPartitioning, outputOrdering): (Partitioning, Seq[SortOrder]) = {
@@ -492,6 +494,11 @@ case class FileSourceScanExec(
       } else {
         Nil
       }
+
+      if (conf.getConf(SQLConf.LEGACY_BUCKETED_TABLE_SCAN_OUTPUT_ORDERING) &&
+        !hasPartitionsAvailableAtRunTime) {
+        logInfo(s"bucketed scan skewed: ${isBucketSkew}")
+      }
       (partitioning, sortOrder)
     } else {
       (UnknownPartitioning(0), Nil)
@@ -536,7 +543,11 @@ case class FileSourceScanExec(
     val withSelectedBucketsCount = relation.bucketSpec.map { spec =>
       val bucketedKey = "Bucketed"
       val withBucketedScanStatus = if (bucketedScan) {
-        metadata + (bucketedKey -> s"true")
+        if (_isBucketSkew.nonEmpty && _isBucketSkew.get) {
+          metadata + (bucketedKey -> s"true, potentialBucketSkewed")
+        } else {
+          metadata + (bucketedKey -> s"true")
+        }
       } else if (!relation.sparkSession.sessionState.conf.bucketingEnabled) {
         metadata + (bucketedKey -> "false (disabled by configuration)")
       } else if (disableBucketedScan) {

diff --git a/...core/src/main/scala/org/apache/spark/sql/execution/adaptive/CustomShuffleReaderExec.scala b/...core/src/main/scala/org/apache/spark/sql/execution/adaptive/CustomShuffleReaderExec.scala
@@ -27,6 +27,7 @@ import org.apache.spark.sql.execution._
 import org.apache.spark.sql.execution.exchange.{ReusedExchangeExec, ShuffleExchangeLike}
 import org.apache.spark.sql.execution.metric.{SQLMetric, SQLMetrics}
 import org.apache.spark.sql.vectorized.ColumnarBatch
+import org.apache.spark.util.Utils
 
 /**
  * A wrapper of shuffle query stage, which follows the given partition arrangement.
@@ -84,11 +85,19 @@ case class CustomShuffleReaderExec private(
     } else if (hasCoalescedPartition && hasSkewedPartition) {
       "coalesced and skewed"
     } else if (hasCoalescedPartition) {
-      "coalesced"
+      if (isPotentialSkewed) {
+        "coalesced and potentialSkewed"
+      } else {
+        "coalesced"
+      }
     } else if (hasSkewedPartition) {
       "skewed"
     } else {
-      ""
+      if (isPotentialSkewed) {
+        "potentialSkewed"
+      } else {
+        ""
+      }
     }
     Iterator(desc)
   }
@@ -102,6 +111,13 @@ case class CustomShuffleReaderExec private(
   def isLocalReader: Boolean =
     partitionSpecs.exists(_.isInstanceOf[PartialMapperPartitionSpec])
 
+  lazy val isPotentialSkewed: Boolean = {
+    !hasSkewedPartition &&
+      partitionDataSizes.nonEmpty &&
+      SkewHandlingUtil.isSkewed(partitionDataSizes.get.max,
+        Utils.median(partitionDataSizes.get.toArray, false), conf)
+  }
+
   private def shuffleStage = child match {
     case stage: ShuffleQueryStageExec => Some(stage)
     case _ => None