HDFS-4911. Reduce PeerCache timeout to be commensurate with dfs.datanode.socket.reuse.keepalive (cmccabe)

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1565438 13f79535-47bb-0310-9956-ffa450edef68

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

@@ -22,6 +22,9 @@ Release 2.4.0 - UNRELEASED
 
     HDFS-5746.  Add ShortCircuitSharedMemorySegment (cmccabe)
 
+    HDFS-4911.  Reduce PeerCache timeout to be commensurate with
+    dfs.datanode.socket.reuse.keepalive (cmccabe)
+
   OPTIMIZATIONS
 
     HDFS-5790. LeaseManager.findPath is very slow when many leases need recovery

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

@@ -86,7 +86,7 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
   public static final int     DFS_CLIENT_RETRY_MAX_ATTEMPTS_DEFAULT = 10;
   
   public static final String  DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_KEY = "dfs.client.socketcache.expiryMsec";
-  public static final long    DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_DEFAULT = 2 * 60 * 1000;
+  public static final long    DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_DEFAULT = 3000;
   public static final String  DFS_CLIENT_WRITE_EXCLUDE_NODES_CACHE_EXPIRY_INTERVAL = "dfs.client.write.exclude.nodes.cache.expiry.interval.millis";
   public static final long    DFS_CLIENT_WRITE_EXCLUDE_NODES_CACHE_EXPIRY_INTERVAL_DEFAULT = 10 * 60 * 1000; // 10 minutes, in ms
   public static final String  DFS_NAMENODE_BACKUP_ADDRESS_KEY = "dfs.namenode.backup.address";
@@ -215,7 +215,7 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
   public static final String  DFS_DATANODE_SYNCONCLOSE_KEY = "dfs.datanode.synconclose";
   public static final boolean DFS_DATANODE_SYNCONCLOSE_DEFAULT = false;
   public static final String  DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_KEY = "dfs.datanode.socket.reuse.keepalive";
-  public static final int     DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_DEFAULT = 1000;
+  public static final int     DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_DEFAULT = 4000;
 
   public static final String DFS_NAMENODE_DATANODE_REGISTRATION_IP_HOSTNAME_CHECK_KEY = "dfs.namenode.datanode.registration.ip-hostname-check";
   public static final boolean DFS_NAMENODE_DATANODE_REGISTRATION_IP_HOSTNAME_CHECK_DEFAULT = true;

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

@@ -18,6 +18,7 @@
 
 package org.apache.hadoop.hdfs;
 
+import java.io.IOException;
 import java.util.Iterator;
 import java.util.List;
 import java.util.Map.Entry;
@@ -25,6 +26,7 @@ import java.util.Map.Entry;
 import com.google.common.annotations.VisibleForTesting;
 import com.google.common.base.Preconditions;
 import com.google.common.collect.LinkedListMultimap;
+
 import org.apache.commons.logging.Log;
 import org.apache.commons.logging.LogFactory;
 import org.apache.hadoop.hdfs.protocol.DatanodeID;
@@ -118,6 +120,11 @@ class PeerCache {
     return instance;
   }
 
+  @VisibleForTesting
+  public static synchronized void setInstance(int c, long e) {
+    instance = new PeerCache(c, e);
+  }
+
   private boolean isDaemonStarted() {
     return (daemon == null)? false: true;
   }
@@ -171,8 +178,17 @@ class PeerCache {
     while (iter.hasNext()) {
       Value candidate = iter.next();
       iter.remove();
-      if (!candidate.getPeer().isClosed()) {
+      long ageMs = Time.monotonicNow() - candidate.getTime();
-        return candidate.getPeer();
+      Peer peer = candidate.getPeer();
+      if (ageMs >= expiryPeriod) {
+        try {
+          peer.close();
+        } catch (IOException e) {
+          LOG.warn("got IOException closing stale peer " + peer +
+                ", which is " + ageMs + " ms old");
+        }
+      } else if (!peer.isClosed()) {
+        return peer;
       }
     }
     return null;

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

@@ -19,16 +19,19 @@ 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_REUSE_KEEPALIVE_DEFAULT;
 import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_DATANODE_SOCKET_WRITE_TIMEOUT_KEY;
+import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_KEY;
+import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_CLIENT_SOCKET_CACHE_CAPACITY_DEFAULT;
 import static org.junit.Assert.assertEquals;
 import static org.junit.Assert.assertNotNull;
 import static org.junit.Assert.assertTrue;
 import static org.junit.Assert.fail;
 
+import com.google.common.base.Supplier;
+
 import java.io.InputStream;
 import java.io.PrintWriter;
-import java.net.InetSocketAddress;
-import java.net.Socket;
 
 import org.apache.hadoop.conf.Configuration;
 import org.apache.hadoop.fs.FSDataInputStream;
@@ -37,10 +40,8 @@ import org.apache.hadoop.fs.Path;
 import org.apache.hadoop.hdfs.MiniDFSCluster.DataNodeProperties;
 import org.apache.hadoop.hdfs.net.Peer;
 import org.apache.hadoop.hdfs.server.datanode.DataNode;
-import org.apache.hadoop.hdfs.server.datanode.DataNodeTestUtils;
-import org.apache.hadoop.hdfs.server.protocol.DatanodeRegistration;
 import org.apache.hadoop.io.IOUtils;
-import org.apache.hadoop.net.NetUtils;
+import org.apache.hadoop.test.GenericTestUtils;
 import org.apache.hadoop.util.ReflectionUtils;
 import org.junit.After;
 import org.junit.Before;
@@ -51,10 +52,7 @@ import com.google.common.io.NullOutputStream;
 public class TestDataTransferKeepalive {
   Configuration conf = new HdfsConfiguration();
   private MiniDFSCluster cluster;
-  private FileSystem fs;
-  private InetSocketAddress dnAddr;
   private DataNode dn;
-  private DFSClient dfsClient;
   private static Path TEST_FILE = new Path("/test");
   
   private static final int KEEPALIVE_TIMEOUT = 1000;
@@ -69,15 +67,7 @@ public class TestDataTransferKeepalive {
     
     cluster = new MiniDFSCluster.Builder(conf)
       .numDataNodes(1).build();
-    fs = cluster.getFileSystem();
-    dfsClient = ((DistributedFileSystem)fs).dfs;
-    dfsClient.peerCache.clear();
-
-    String poolId = cluster.getNamesystem().getBlockPoolId();
     dn = cluster.getDataNodes().get(0);
-    DatanodeRegistration dnReg = DataNodeTestUtils.getDNRegistrationForBP(
-        dn, poolId);
-    dnAddr = NetUtils.createSocketAddr(dnReg.getXferAddr());
   }
   
   @After
@@ -90,34 +80,86 @@ public class TestDataTransferKeepalive {
    * its configured keepalive timeout.
    */
   @Test(timeout=30000)
-  public void testKeepaliveTimeouts() throws Exception {
+  public void testDatanodeRespectsKeepAliveTimeout() throws Exception {
+    Configuration clientConf = new Configuration(conf);
+    // Set a client socket cache expiry time much longer than 
+    // the datanode-side expiration time.
+    final long CLIENT_EXPIRY_MS = 60000L;
+    clientConf.setLong(DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_KEY, CLIENT_EXPIRY_MS);
+    PeerCache.setInstance(DFS_CLIENT_SOCKET_CACHE_CAPACITY_DEFAULT, CLIENT_EXPIRY_MS);
+    DistributedFileSystem fs =
+        (DistributedFileSystem)FileSystem.get(cluster.getURI(),
+            clientConf);
+
     DFSTestUtil.createFile(fs, TEST_FILE, 1L, (short)1, 0L);
 
     // Clients that write aren't currently re-used.
-    assertEquals(0, dfsClient.peerCache.size());
+    assertEquals(0, fs.dfs.peerCache.size());
     assertXceiverCount(0);
 
     // Reads the file, so we should get a
     // cached socket, and should have an xceiver on the other side.
     DFSTestUtil.readFile(fs, TEST_FILE);
-    assertEquals(1, dfsClient.peerCache.size());
+    assertEquals(1, fs.dfs.peerCache.size());
     assertXceiverCount(1);
 
     // Sleep for a bit longer than the keepalive timeout
     // and make sure the xceiver died.
-    Thread.sleep(KEEPALIVE_TIMEOUT * 2);
+    Thread.sleep(DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_DEFAULT + 1);
     assertXceiverCount(0);
     
     // The socket is still in the cache, because we don't
     // notice that it's closed until we try to read
     // from it again.
-    assertEquals(1, dfsClient.peerCache.size());
+    assertEquals(1, fs.dfs.peerCache.size());
     
     // Take it out of the cache - reading should
     // give an EOF.
-    Peer peer = dfsClient.peerCache.get(dn.getDatanodeId(), false);
+    Peer peer = fs.dfs.peerCache.get(dn.getDatanodeId(), false);
     assertNotNull(peer);
     assertEquals(-1, peer.getInputStream().read());
+    PeerCache.setInstance(DFS_CLIENT_SOCKET_CACHE_CAPACITY_DEFAULT,
+        DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_DEFAULT);
+  }
+
+  /**
+   * Test that the client respects its keepalive timeout.
+   */
+  @Test(timeout=30000)
+  public void testClientResponsesKeepAliveTimeout() throws Exception {
+    Configuration clientConf = new Configuration(conf);
+    // Set a client socket cache expiry time much shorter than 
+    // the datanode-side expiration time.
+    final long CLIENT_EXPIRY_MS = 10L;
+    clientConf.setLong(DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_KEY, CLIENT_EXPIRY_MS);
+    PeerCache.setInstance(DFS_CLIENT_SOCKET_CACHE_CAPACITY_DEFAULT, CLIENT_EXPIRY_MS);
+    DistributedFileSystem fs =
+        (DistributedFileSystem)FileSystem.get(cluster.getURI(),
+            clientConf);
+
+    DFSTestUtil.createFile(fs, TEST_FILE, 1L, (short)1, 0L);
+
+    // Clients that write aren't currently re-used.
+    assertEquals(0, fs.dfs.peerCache.size());
+    assertXceiverCount(0);
+
+    // Reads the file, so we should get a
+    // cached socket, and should have an xceiver on the other side.
+    DFSTestUtil.readFile(fs, TEST_FILE);
+    assertEquals(1, fs.dfs.peerCache.size());
+    assertXceiverCount(1);
+
+    // Sleep for a bit longer than the client keepalive timeout.
+    Thread.sleep(CLIENT_EXPIRY_MS + 1);
+    
+    // Taking out a peer which is expired should give a null.
+    Peer peer = fs.dfs.peerCache.get(dn.getDatanodeId(), false);
+    assertTrue(peer == null);
+
+    // The socket cache is now empty.
+    assertEquals(0, fs.dfs.peerCache.size());
+    PeerCache.setInstance(DFS_CLIENT_SOCKET_CACHE_CAPACITY_DEFAULT,
+        DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_DEFAULT);
   }
 
   /**
@@ -125,8 +167,17 @@ public class TestDataTransferKeepalive {
    * read bytes off the stream quickly. The datanode should time out sending the
    * chunks and the transceiver should die, even if it has a long keepalive.
    */
-  @Test(timeout=30000)
+  @Test(timeout=300000)
   public void testSlowReader() throws Exception {
+    // Set a client socket cache expiry time much longer than 
+    // the datanode-side expiration time.
+    final long CLIENT_EXPIRY_MS = 600000L;
+    Configuration clientConf = new Configuration(conf);
+    clientConf.setLong(DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_KEY, CLIENT_EXPIRY_MS);
+    PeerCache.setInstance(DFS_CLIENT_SOCKET_CACHE_CAPACITY_DEFAULT, CLIENT_EXPIRY_MS);
+    DistributedFileSystem fs =
+        (DistributedFileSystem)FileSystem.get(cluster.getURI(),
+            clientConf);
     // Restart the DN with a shorter write timeout.
     DataNodeProperties props = cluster.stopDataNode(0);
     props.conf.setInt(DFS_DATANODE_SOCKET_WRITE_TIMEOUT_KEY,
@@ -134,38 +185,31 @@ public class TestDataTransferKeepalive {
     props.conf.setInt(DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_KEY,
         120000);
     assertTrue(cluster.restartDataNode(props, true));
+    dn = cluster.getDataNodes().get(0);
     // Wait for heartbeats to avoid a startup race where we
     // try to write the block while the DN is still starting.
     cluster.triggerHeartbeats();
     
-    dn = cluster.getDataNodes().get(0);
-    
     DFSTestUtil.createFile(fs, TEST_FILE, 1024*1024*8L, (short)1, 0L);
     FSDataInputStream stm = fs.open(TEST_FILE);
-    try {
+    stm.read();
-      stm.read();
+    assertXceiverCount(1);
-      assertXceiverCount(1);
 
-      // Poll for 0 running xceivers.  Allow up to 5 seconds for some slack.
+    GenericTestUtils.waitFor(new Supplier<Boolean>() {
-      long totalSleepTime = 0;
+      public Boolean get() {
-      long sleepTime = WRITE_TIMEOUT + 100;
+        // DN should time out in sendChunks, and this should force
-      while (getXceiverCountWithoutServer() > 0 && totalSleepTime < 5000) {
+        // the xceiver to exit.
-        Thread.sleep(sleepTime);
+        return getXceiverCountWithoutServer() == 0;
-        totalSleepTime += sleepTime;
-        sleepTime = 100;
       }
+    }, 500, 50000);
 
-      // DN should time out in sendChunks, and this should force
+    IOUtils.closeStream(stm);
-      // the xceiver to exit.
-      assertXceiverCount(0);
-    } finally {
-      IOUtils.closeStream(stm);
-    }
   }
   
   @Test(timeout=30000)
   public void testManyClosedSocketsInCache() throws Exception {
     // Make a small file
+    DistributedFileSystem fs = cluster.getFileSystem();
     DFSTestUtil.createFile(fs, TEST_FILE, 1L, (short)1, 0L);
 
     // Insert a bunch of dead sockets in the cache, by opening