[SPARK-11276][CORE] SizeEstimator prevents class unloading

The SizeEstimator keeps a cache of ClassInfos but this cache uses Class objects as keys.
Which results in strong references to the Class objects. If these classes are dynamically created
this prevents the corresponding ClassLoader from being GCed. Leading to PermGen exhaustion.

We use a Map with WeakKeys to prevent this issue.

Author: Sem Mulder <sem.mulder@site2mobile.com>

Closes #9244 from SemMulder/fix-sizeestimator-classunloading.

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

@@ -17,6 +17,8 @@
 
 package org.apache.spark.util
 
+import com.google.common.collect.MapMaker
+
 import java.lang.management.ManagementFactory
 import java.lang.reflect.{Field, Modifier}
 import java.util.{IdentityHashMap, Random}
@@ -29,7 +31,6 @@ import org.apache.spark.Logging
 import org.apache.spark.annotation.DeveloperApi
 import org.apache.spark.util.collection.OpenHashSet
 
-
 /**
  * :: DeveloperApi ::
  * Estimates the sizes of Java objects (number of bytes of memory they occupy), for use in
@@ -73,7 +74,8 @@ object SizeEstimator extends Logging {
   private val ALIGN_SIZE = 8
 
   // A cache of ClassInfo objects for each class
-  private val classInfos = new ConcurrentHashMap[Class[_], ClassInfo]
+  // We use weakKeys to allow GC of dynamically created classes
+  private val classInfos = new MapMaker().weakKeys().makeMap[Class[_], ClassInfo]()
 
   // Object and pointer sizes are arch dependent
   private var is64bit = false