HDFS-3376. DFSClient fails to make connection to DN if there are many unusable cached sockets. Contributed by Todd Lipcon.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1335114 13f79535-47bb-0310-9956-ffa450edef68

hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt

@@ -485,6 +485,9 @@ Release 2.0.0 - UNRELEASED
     HDFS-3357. DataXceiver reads from client socket with incorrect/no timeout
     (todd)
 
+    HDFS-3376. DFSClient fails to make connection to DN if there are many
+    unusable cached sockets (todd)
+
   BREAKDOWN OF HDFS-1623 SUBTASKS
 
     HDFS-2179. Add fencing framework and mechanisms for NameNode HA. (todd)

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/DFSInputStream.java

@@ -794,7 +794,13 @@ public class DFSInputStream extends FSInputStream {
     // Allow retry since there is no way of knowing whether the cached socket
     // is good until we actually use it.
     for (int retries = 0; retries <= nCachedConnRetry && fromCache; ++retries) {
-      Socket sock = socketCache.get(dnAddr);
+      Socket sock = null;
+      // Don't use the cache on the last attempt - it's possible that there
+      // are arbitrarily many unusable sockets in the cache, but we don't
+      // want to fail the read.
+      if (retries < nCachedConnRetry) {
+        sock = socketCache.get(dnAddr);
+      }
       if (sock == null) {
         fromCache = false;
 

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDataTransferKeepalive.java

@@ -17,10 +17,12 @@
  */
 package org.apache.hadoop.hdfs;
 
+import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_CLIENT_MAX_BLOCK_ACQUIRE_FAILURES_KEY;
 import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_KEY;
 import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_DATANODE_SOCKET_WRITE_TIMEOUT_KEY;
 import static org.junit.Assert.*;
 
+import java.io.InputStream;
 import java.io.PrintWriter;
 import java.net.InetSocketAddress;
 import java.net.Socket;
@@ -40,6 +42,8 @@ import org.junit.After;
 import org.junit.Before;
 import org.junit.Test;
 
+import com.google.common.io.NullOutputStream;
+
 public class TestDataTransferKeepalive {
   Configuration conf = new HdfsConfiguration();
   private MiniDFSCluster cluster;
@@ -56,6 +60,8 @@ public class TestDataTransferKeepalive {
   public void setup() throws Exception {
     conf.setInt(DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_KEY,
         KEEPALIVE_TIMEOUT);
+    conf.setInt(DFS_CLIENT_MAX_BLOCK_ACQUIRE_FAILURES_KEY,
+        0);
     
     cluster = new MiniDFSCluster.Builder(conf)
       .numDataNodes(1).build();
@@ -143,6 +149,40 @@ public class TestDataTransferKeepalive {
       IOUtils.closeStream(stm);
     }
   }
+  
+  @Test(timeout=30000)
+  public void testManyClosedSocketsInCache() throws Exception {
+    // Make a small file
+    DFSTestUtil.createFile(fs, TEST_FILE, 1L, (short)1, 0L);
+
+    // Insert a bunch of dead sockets in the cache, by opening
+    // many streams concurrently, reading all of the data,
+    // and then closing them.
+    InputStream[] stms = new InputStream[5];
+    try {
+      for (int i = 0; i < stms.length; i++) {
+        stms[i] = fs.open(TEST_FILE);
+      }
+      for (InputStream stm : stms) {
+        IOUtils.copyBytes(stm, new NullOutputStream(), 1024);
+      }
+    } finally {
+      IOUtils.cleanup(null, stms);
+    }
+    
+    DFSClient client = ((DistributedFileSystem)fs).dfs;
+    assertEquals(5, client.socketCache.size());
+    
+    // Let all the xceivers timeout
+    Thread.sleep(1500);
+    assertXceiverCount(0);
+
+    // Client side still has the sockets cached
+    assertEquals(5, client.socketCache.size());
+
+    // Reading should not throw an exception.
+    DFSTestUtil.readFile(fs, TEST_FILE);
+  }
 
   private void assertXceiverCount(int expected) {
     // Subtract 1, since the DataXceiverServer