[SPARK-13747][SQL] Fix concurrent executions in ForkJoinPool for SQL

Calling `Await.result` will allow other tasks to be run on the same thread when using ForkJoinPool. However, SQL uses a `ThreadLocal` execution id to trace Spark jobs launched by a query, which doesn't work perfectly in ForkJoinPool.

This PR just uses `Awaitable.result` instead to  prevent ForkJoinPool from running other tasks in the current waiting thread.

Jenkins

Author: Shixiong Zhu <[email protected]>

Closes #15520 from zsxwing/SPARK-13747.

(cherry picked from commit 7ac70e7)
Signed-off-by: Shixiong Zhu <[email protected]>

Author: zsxwing

core/src/main/scala/org/apache/spark/util/ThreadUtils.scala

@@ -194,4 +194,25 @@ private[spark] object ThreadUtils {
         throw new SparkException("Exception thrown in awaitResult: ", t)
     }
   }
+
+  /**
+   * Calls [[Awaitable.result]] directly to avoid using `ForkJoinPool`'s `BlockingContext`, wraps
+   * and re-throws any exceptions with nice stack track.
+   *
+   * Codes running in the user's thread may be in a thread of Scala ForkJoinPool. As concurrent
+   * executions in ForkJoinPool may see some [[ThreadLocal]] value unexpectedly, this method
+   * basically prevents ForkJoinPool from running other tasks in the current waiting thread.
+   */
+  @throws(classOf[SparkException])
+  def awaitResultInForkJoinSafely[T](awaitable: Awaitable[T], atMost: Duration): T = {
+    try {
+      // `awaitPermission` is not actually used anywhere so it's safe to pass in null here.
+      // See SPARK-13747.
+      val awaitPermission = null.asInstanceOf[scala.concurrent.CanAwait]
+      awaitable.result(Duration.Inf)(awaitPermission)
+    } catch {
+      case NonFatal(t) =>
+        throw new SparkException("Exception thrown in awaitResult: ", t)
+    }
+  }
 }

scalastyle-config.xml

@@ -200,6 +200,7 @@ This file is divided into 3 sections:
       // scalastyle:off awaitresult
       Await.result(...)
       // scalastyle:on awaitresult
+      If your codes use ThreadLocal and may run in threads created by the user, use ThreadUtils.awaitResultInForkJoinSafely instead.
     ]]></customMessage>
   </check>
 

sql/core/src/main/scala/org/apache/spark/sql/execution/SparkPlan.scala

@@ -166,7 +166,7 @@ abstract class SparkPlan extends QueryPlan[SparkPlan] with Logging with Serializ
   protected def waitForSubqueries(): Unit = synchronized {
     // fill in the result of subqueries
     subqueryResults.foreach { case (e, futureResult) =>
-      val rows = ThreadUtils.awaitResult(futureResult, Duration.Inf)
+      val rows = ThreadUtils.awaitResultInForkJoinSafely(futureResult, Duration.Inf)
       if (rows.length > 1) {
         sys.error(s"more than one row returned by a subquery used as an expression:\n${e.plan}")
       }

sql/core/src/main/scala/org/apache/spark/sql/execution/exchange/BroadcastExchangeExec.scala

@@ -117,7 +117,8 @@ case class BroadcastExchangeExec(
   }
 
   override protected[sql] def doExecuteBroadcast[T](): broadcast.Broadcast[T] = {
-    ThreadUtils.awaitResult(relationFuture, timeout).asInstanceOf[broadcast.Broadcast[T]]
+    ThreadUtils.awaitResultInForkJoinSafely(relationFuture, timeout)
+      .asInstanceOf[broadcast.Broadcast[T]]
   }
 }