[SPARK-21343] Refine the document for spark.reducer.maxReqSizeShuffleToMem.

## What changes were proposed in this pull request?

In current code, reducer can break the old shuffle service when `spark.reducer.maxReqSizeShuffleToMem` is enabled. Let's refine document.

Author: jinxing <jinxing6042@126.com>

Closes #18566 from jinxing64/SPARK-21343.

core/src/main/scala/org/apache/spark/internal/config/package.scala

@@ -323,9 +323,11 @@ package object config {
 
   private[spark] val REDUCER_MAX_REQ_SIZE_SHUFFLE_TO_MEM =
     ConfigBuilder("spark.reducer.maxReqSizeShuffleToMem")
-      .internal()
       .doc("The blocks of a shuffle request will be fetched to disk when size of the request is " +
-        "above this threshold. This is to avoid a giant request takes too much memory.")
+        "above this threshold. This is to avoid a giant request takes too much memory. We can " +
+        "enable this config by setting a specific value(e.g. 200m). Note that this config can " +
+        "be enabled only when the shuffle shuffle service is newer than Spark-2.2 or the shuffle" +
+        " service is disabled.")
       .bytesConf(ByteUnit.BYTE)
       .createWithDefault(Long.MaxValue)
 

docs/configuration.md

@@ -528,6 +528,16 @@ Apart from these, the following properties are also available, and may be useful
     By allowing it to limit the number of fetch requests, this scenario can be mitigated.
   </td>
 </tr>
+<tr>
+  <td><code>spark.reducer.maxReqSizeShuffleToMem</code></td>
+  <td>Long.MaxValue</td>
+  <td>
+    The blocks of a shuffle request will be fetched to disk when size of the request is above
+    this threshold. This is to avoid a giant request takes too much memory. We can enable this
+    config by setting a specific value(e.g. 200m). Note that this config can be enabled only when
+    the shuffle shuffle service is newer than Spark-2.2 or the shuffle service is disabled.
+  </td>
+</tr>
 <tr>
   <td><code>spark.shuffle.compress</code></td>
   <td>true</td>