cap numPartsToTry

core/src/main/scala/org/apache/spark/rdd/AsyncRDDActions.scala

@@ -78,14 +78,15 @@ class AsyncRDDActions[T: ClassTag](self: RDD[T]) extends Serializable with Loggi
         // greater than totalParts because we actually cap it at totalParts in runJob.
         var numPartsToTry = 1
         if (partsScanned > 0) {
-          // If we didn't find any rows after the first iteration, just try all partitions next.
+          // If we didn't find any rows after the previous iteration, quadruple and retry.
           // Otherwise, interpolate the number of partitions we need to try, but overestimate it
-          // by 50%.
+          // by 50%. We also cap the estimation in the end.
           if (results.size == 0) {
-            numPartsToTry = totalParts - 1
+            numPartsToTry = totalParts * 4
           } else {
             // the left side of max is >=1 whenever partsScanned >= 2
             numPartsToTry = ((1.5 * num * partsScanned / results.size).toInt - partsScanned) max 1
+            numPartsToTry = numPartsToTry min (totalParts * 4) 
           }
         }
 

core/src/main/scala/org/apache/spark/rdd/RDD.scala

@@ -1079,13 +1079,15 @@ abstract class RDD[T: ClassTag](
       // greater than totalParts because we actually cap it at totalParts in runJob.
       var numPartsToTry = 1
       if (partsScanned > 0) {
-        // If we didn't find any rows after the previous iteration, quadruple and retry.  Otherwise,
+        // If we didn't find any rows after the previous iteration, quadruple and retry. Otherwise,
         // interpolate the number of partitions we need to try, but overestimate it by 50%.
+        // We also cap the estimation in the end.
         if (buf.size == 0) {
           numPartsToTry = partsScanned * 4
         } else {
           // the left side of max is >=1 whenever partsScanned >= 2
           numPartsToTry = ((1.5 * num * partsScanned / buf.size).toInt - partsScanned) max 1
+          numPartsToTry = numPartsToTry min (partsScanned * 4) 
         }
       }
 

python/pyspark/rdd.py

@@ -1070,11 +1070,13 @@ def take(self, num):
                 # If we didn't find any rows after the previous iteration,
                 # quadruple and retry.  Otherwise, interpolate the number of
                 # partitions we need to try, but overestimate it by 50%.
+                # We also cap the estimation in the end.
                 if len(items) == 0:
                     numPartsToTry = partsScanned * 4
                 else:
                     #the first paramter of max is >=1 whenever partsScanned >= 2
                     numPartsToTry = max(int(1.5 * num * partsScanned / len(items)) - partsScanned, 1)
+                    numPartsToTry = min(numPartsToTry, partsScanned * 4)
 
             left = num - len(items)