[SPARK-7490] [CORE] [Minor] MapOutputTracker.deserializeMapStatuses: close input streams

Evan Jones · srowen · commit 25889d8d9709 · 2015-05-08T22:00:39.000+01:00
GZIPInputStream allocates native memory that is not freed until close() or when the finalizer runs. It is best to close() these streams explicitly. stephenh made the same change for serializeMapStatuses in commit b0d884f. This is the same change for deserialize. (I ran the unit test suite! it seems to have passed. I did not make a JIRA since this seems "trivial", and the guidelines suggest it is not required for trivial changes) Author: Evan Jones <ejones@twitter.com> Closes #5982 from evanj/master and squashes the following commits: 0d76e85 [Evan Jones] [CORE] MapOutputTracker.deserializeMapStatuses: close input streams
diff --git a/core/src/main/scala/org/apache/spark/MapOutputTracker.scala b/core/src/main/scala/org/apache/spark/MapOutputTracker.scala
@@ -367,7 +367,11 @@ private[spark] object MapOutputTracker extends Logging {
   // Opposite of serializeMapStatuses.
   def deserializeMapStatuses(bytes: Array[Byte]): Array[MapStatus] = {
     val objIn = new ObjectInputStream(new GZIPInputStream(new ByteArrayInputStream(bytes)))
-    objIn.readObject().asInstanceOf[Array[MapStatus]]
+    Utils.tryWithSafeFinally {
+      objIn.readObject().asInstanceOf[Array[MapStatus]]
+    } {
+      objIn.close()
+    }
   }
 
   // Convert an array of MapStatuses to locations and sizes for a given reduce ID. If
