Fix spark.io.compression.codec and change default codec to LZF

core/src/main/scala/org/apache/spark/io/CompressionCodec.scala

@@ -39,17 +39,13 @@ trait CompressionCodec {
 private[spark] object CompressionCodec {
 
   def createCodec(): CompressionCodec = {
-    // Set the default codec to Snappy since the LZF implementation initializes a pretty large
-    // buffer for every stream, which results in a lot of memory overhead when the number of
-    // shuffle reduce buckets are large.
-    createCodec(classOf[SnappyCompressionCodec].getName)
+    createCodec(System.getProperty(
+      "spark.io.compression.codec", classOf[LZFCompressionCodec].getName))
   }
 
   def createCodec(codecName: String): CompressionCodec = {
-    Class.forName(
-      System.getProperty("spark.io.compression.codec", codecName),
-      true,
-      Thread.currentThread.getContextClassLoader).newInstance().asInstanceOf[CompressionCodec]
+    Class.forName(codecName, true, Thread.currentThread.getContextClassLoader)
+      .newInstance().asInstanceOf[CompressionCodec]
   }
 }
 

docs/configuration.md

@@ -147,7 +147,7 @@ Apart from these, the following properties are also available, and may be useful
 </tr>
 <tr>
   <td>spark.io.compression.codec</td>
-  <td>org.apache.spark.io.<br />SnappyCompressionCodec</td>
+  <td>org.apache.spark.io.<br />LZFCompressionCodec</td>
   <td>
     The compression codec class to use for various compressions. By default, Spark provides two
     codecs: <code>org.apache.spark.io.LZFCompressionCodec</code> and <code>org.apache.spark.io.SnappyCompressionCodec</code>.