svn commit: r727277 - in /hadoop/hbase/trunk: CHANGES.txt src/java/org/apache/hadoop/hbase/ipc/HBaseServer.java

[st...@apache.org]

Author: stack
Date: Tue Dec 16 20:47:14 2008
New Revision: 727277

URL: http://svn.apache.org/viewvc?rev=727277&view=rev
Log:
HBASE-1053 bring recent rpc changes down from hadoop

Modified:
    hadoop/hbase/trunk/CHANGES.txt
    hadoop/hbase/trunk/src/java/org/apache/hadoop/hbase/ipc/HBaseServer.java

Modified: hadoop/hbase/trunk/CHANGES.txt
URL: http://svn.apache.org/viewvc/hadoop/hbase/trunk/CHANGES.txt?rev=727277&r1=727276&r2=727277&view=diff
==============================================================================
--- hadoop/hbase/trunk/CHANGES.txt (original)
+++ hadoop/hbase/trunk/CHANGES.txt Tue Dec 16 20:47:14 2008
@@ -175,6 +175,7 @@
                outstanding seqnum is 162297053 fr om region -ROOT-,,0
    HBASE-1055  Better vm stats on startup
    HBASE-1065  Minor logging improvements in the master
+   HBASE-1053  bring recent rpc changes down from hadoop
 
   NEW FEATURES
    HBASE-875   Use MurmurHash instead of JenkinsHash [in bloomfilters]

Modified: hadoop/hbase/trunk/src/java/org/apache/hadoop/hbase/ipc/HBaseServer.java
URL: http://svn.apache.org/viewvc/hadoop/hbase/trunk/src/java/org/apache/hadoop/hbase/ipc/HBaseServer.java?rev=727277&r1=727276&r2=727277&view=diff
==============================================================================
--- hadoop/hbase/trunk/src/java/org/apache/hadoop/hbase/ipc/HBaseServer.java (original)
+++ hadoop/hbase/trunk/src/java/org/apache/hadoop/hbase/ipc/HBaseServer.java Tue Dec 16 20:47:14 2008
@@ -37,6 +37,8 @@
 import java.nio.channels.Selector;
 import java.nio.channels.ServerSocketChannel;
 import java.nio.channels.SocketChannel;
+import java.nio.channels.ReadableByteChannel;
+import java.nio.channels.WritableByteChannel;
 import java.util.ArrayList;
 import java.util.Collections;
 import java.util.Iterator;
@@ -587,7 +589,7 @@
           //
           // Send as much data as we can in the non-blocking fashion
           //
-          int numBytes = channel.write(call.response);
+          int numBytes = channelWrite(channel, call.response);
           if (numBytes < 0) {
             return true;
           }
@@ -762,7 +764,7 @@
          */    
         int count = -1;
         if (dataLengthBuffer.remaining() > 0) {
-          count = channel.read(dataLengthBuffer);       
+          count = channelRead(channel, dataLengthBuffer);       
           if (count < 0 || dataLengthBuffer.remaining() > 0) 
             return count;
         }
@@ -770,7 +772,7 @@
         if (!versionRead) {
           //Every connection is expected to send the header.
           ByteBuffer versionBuffer = ByteBuffer.allocate(1);
-          count = channel.read(versionBuffer);
+          count = channelRead(channel, versionBuffer);
           if (count <= 0) {
             return count;
           }
@@ -802,7 +804,7 @@
           incRpcCount();  // Increment the rpc count
         }
         
-        count = channel.read(data);
+        count = channelRead(channel, data);
         
         if (data.remaining() == 0) {
           dataLengthBuffer.clear();
@@ -1058,4 +1060,80 @@
     return callQueue.size();
   }
   
+  
+  /**
+   * When the read or write buffer size is larger than this limit, i/o will be 
+   * done in chunks of this size. Most RPC requests and responses would be
+   * be smaller.
+   */
+  private static int NIO_BUFFER_LIMIT = 8*1024; //should not be more than 64KB.
+  
+  /**
+   * This is a wrapper around {@link WritableByteChannel#write(ByteBuffer)}.
+   * If the amount of data is large, it writes to channel in smaller chunks. 
+   * This is to avoid jdk from creating many direct buffers as the size of 
+   * buffer increases. This also minimizes extra copies in NIO layer
+   * as a result of multiple write operations required to write a large 
+   * buffer.  
+   *
+   * @see WritableByteChannel#write(ByteBuffer)
+   */
+  private static int channelWrite(WritableByteChannel channel, 
+                                  ByteBuffer buffer) throws IOException {
+    
+    return (buffer.remaining() <= NIO_BUFFER_LIMIT) ?
+           channel.write(buffer) : channelIO(null, channel, buffer);
+  }
+  
+  
+  /**
+   * This is a wrapper around {@link ReadableByteChannel#read(ByteBuffer)}.
+   * If the amount of data is large, it writes to channel in smaller chunks. 
+   * This is to avoid jdk from creating many direct buffers as the size of 
+   * ByteBuffer increases. There should not be any performance degredation.
+   * 
+   * @see ReadableByteChannel#read(ByteBuffer)
+   */
+  private static int channelRead(ReadableByteChannel channel, 
+                                 ByteBuffer buffer) throws IOException {
+    
+    return (buffer.remaining() <= NIO_BUFFER_LIMIT) ?
+           channel.read(buffer) : channelIO(channel, null, buffer);
+  }
+  
+  /**
+   * Helper for {@link #channelRead(ReadableByteChannel, ByteBuffer)}
+   * and {@link #channelWrite(WritableByteChannel, ByteBuffer)}. Only
+   * one of readCh or writeCh should be non-null.
+   * 
+   * @see #channelRead(ReadableByteChannel, ByteBuffer)
+   * @see #channelWrite(WritableByteChannel, ByteBuffer)
+   */
+  private static int channelIO(ReadableByteChannel readCh, 
+                               WritableByteChannel writeCh,
+                               ByteBuffer buf) throws IOException {
+    
+    int originalLimit = buf.limit();
+    int initialRemaining = buf.remaining();
+    int ret = 0;
+    
+    while (buf.remaining() > 0) {
+      try {
+        int ioSize = Math.min(buf.remaining(), NIO_BUFFER_LIMIT);
+        buf.limit(buf.position() + ioSize);
+        
+        ret = (readCh == null) ? writeCh.write(buf) : readCh.read(buf); 
+        
+        if (ret < ioSize) {
+          break;
+        }
+
+      } finally {
+        buf.limit(originalLimit);        
+      }
+    }
+
+    int nBytes = initialRemaining - buf.remaining(); 
+    return (nBytes > 0) ? nBytes : ret;
+  }      
 }