HDFS-1580. Add interface for generic Write Ahead Logging mechanisms. Contributed by Ivan Kelly.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1210602 13f79535-47bb-0310-9956-ffa450edef68

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

@@ -72,6 +72,9 @@ Trunk (unreleased changes)
 									Move the support for multiple protocols to lower layer so that Writable,
 								  PB and Avro can all use it (Sanjay)
 
+    HDFS-1580. Add interface for generic Write Ahead Logging mechanisms.
+    (Ivan Kelly via jitendra)
+
   OPTIMIZATIONS
     HDFS-2477. Optimize computing the diff between a block report and the
                namenode state. (Tomasz Nykiel via hairong)

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

@@ -161,6 +161,7 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
   public static final String  DFS_NAMENODE_HTTPS_ADDRESS_DEFAULT = "0.0.0.0:" + DFS_NAMENODE_HTTPS_PORT_DEFAULT;
   public static final String  DFS_NAMENODE_NAME_DIR_KEY = "dfs.namenode.name.dir";
   public static final String  DFS_NAMENODE_EDITS_DIR_KEY = "dfs.namenode.edits.dir";
+  public static final String  DFS_NAMENODE_EDITS_PLUGIN_PREFIX = "dfs.namenode.edits.journal-plugin";
   public static final String  DFS_CLIENT_READ_PREFETCH_SIZE_KEY = "dfs.client.read.prefetch.size"; 
   public static final String  DFS_CLIENT_RETRY_WINDOW_BASE= "dfs.client.retry.window.base";
   public static final String  DFS_METRICS_SESSION_ID_KEY = "dfs.metrics.session-id";

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/EditLogBackupInputStream.java

@@ -103,7 +103,7 @@ public void close() throws IOException {
   }
 
   @Override
-  long length() throws IOException {
+  public long length() throws IOException {
     // file size + size of both buffers
     return inner.length();
   }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/EditLogBackupOutputStream.java

@@ -67,12 +67,12 @@ class EditLogBackupOutputStream extends EditLogOutputStream {
   }
   
   @Override // EditLogOutputStream
-  void write(FSEditLogOp op) throws IOException {
+  public void write(FSEditLogOp op) throws IOException {
     doubleBuf.writeOp(op);
  }
 
   @Override
-  void writeRaw(byte[] bytes, int offset, int length) throws IOException {
+  public void writeRaw(byte[] bytes, int offset, int length) throws IOException {
     throw new IOException("Not supported");
   }
 
@@ -80,7 +80,7 @@ void writeRaw(byte[] bytes, int offset, int length) throws IOException {
    * There is no persistent storage. Just clear the buffers.
    */
   @Override // EditLogOutputStream
-  void create() throws IOException {
+  public void create() throws IOException {
     assert doubleBuf.isFlushed() : "previous data is not flushed yet";
     this.doubleBuf = new EditsDoubleBuffer(DEFAULT_BUFFER_SIZE);
   }
@@ -106,7 +106,7 @@ public void abort() throws IOException {
   }
 
   @Override // EditLogOutputStream
-  void setReadyToFlush() throws IOException {
+  public void setReadyToFlush() throws IOException {
     doubleBuf.setReadyToFlush();
   }
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/EditLogFileInputStream.java

@@ -127,7 +127,7 @@ public void close() throws IOException {
   }
 
   @Override
-  long length() throws IOException {
+  public long length() throws IOException {
     // file size + size of both buffers
     return file.length();
   }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/EditLogFileOutputStream.java

@@ -73,7 +73,7 @@ class EditLogFileOutputStream extends EditLogOutputStream {
 
   /** {@inheritDoc} */
   @Override
-  void write(FSEditLogOp op) throws IOException {
+  public void write(FSEditLogOp op) throws IOException {
     doubleBuf.writeOp(op);
   }
 
@@ -86,7 +86,7 @@ void write(FSEditLogOp op) throws IOException {
    * </ul>
    * */
   @Override
-  void writeRaw(byte[] bytes, int offset, int length) throws IOException {
+  public void writeRaw(byte[] bytes, int offset, int length) throws IOException {
     doubleBuf.writeRaw(bytes, offset, length);
   }
 
@@ -94,7 +94,7 @@ void writeRaw(byte[] bytes, int offset, int length) throws IOException {
    * Create empty edits logs file.
    */
   @Override
-  void create() throws IOException {
+  public void create() throws IOException {
     fc.truncate(0);
     fc.position(0);
     doubleBuf.getCurrentBuf().writeInt(HdfsConstants.LAYOUT_VERSION);
@@ -150,7 +150,7 @@ public void abort() throws IOException {
    * data can be still written to the stream while flushing is performed.
    */
   @Override
-  void setReadyToFlush() throws IOException {
+  public void setReadyToFlush() throws IOException {
     doubleBuf.getCurrentBuf().write(FSEditLogOpCodes.OP_INVALID.getOpCode()); // insert eof marker
     doubleBuf.setReadyToFlush();
   }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/EditLogInputStream.java

@@ -17,6 +17,8 @@
  */
 package org.apache.hadoop.hdfs.server.namenode;
 
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
 import java.io.Closeable;
 import java.io.IOException;
 
@@ -27,7 +29,9 @@
  * It should stream bytes from the storage exactly as they were written
  * into the #{@link EditLogOutputStream}.
  */
-abstract class EditLogInputStream implements JournalStream, Closeable {
+@InterfaceAudience.Private
+@InterfaceStability.Evolving
+public abstract class EditLogInputStream implements JournalStream, Closeable {
   /** 
    * @return the first transaction which will be found in this stream
    */
@@ -74,5 +78,5 @@ abstract class EditLogInputStream implements JournalStream, Closeable {
   /**
    * Return the size of the current edits log.
    */
-  abstract long length() throws IOException;
+  public abstract long length() throws IOException;
 }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/EditLogOutputStream.java

@@ -21,17 +21,21 @@
 
 import static org.apache.hadoop.hdfs.server.common.Util.now;
 
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
 
 /**
  * A generic abstract class to support journaling of edits logs into 
  * a persistent storage.
  */
-abstract class EditLogOutputStream {
+@InterfaceAudience.Private
+@InterfaceStability.Evolving
+public abstract class EditLogOutputStream {
   // these are statistics counters
   private long numSync;        // number of sync(s) to disk
   private long totalTimeSync;  // total time to sync
 
-  EditLogOutputStream() {
+  public EditLogOutputStream() throws IOException {
     numSync = totalTimeSync = 0;
   }
 
@@ -41,7 +45,7 @@ abstract class EditLogOutputStream {
    * @param op operation
    * @throws IOException
    */
-  abstract void write(FSEditLogOp op) throws IOException;
+  abstract public void write(FSEditLogOp op) throws IOException;
 
   /**
    * Write raw data to an edit log. This data should already have
@@ -54,7 +58,7 @@ abstract class EditLogOutputStream {
    * @param length number of bytes to write
    * @throws IOException
    */
-  abstract void writeRaw(byte[] bytes, int offset, int length)
+  abstract public void writeRaw(byte[] bytes, int offset, int length)
       throws IOException;
 
   /**
@@ -62,7 +66,7 @@ abstract void writeRaw(byte[] bytes, int offset, int length)
    * 
    * @throws IOException
    */
-  abstract void create() throws IOException;
+  abstract public void create() throws IOException;
 
   /**
    * Close the journal.
@@ -81,7 +85,7 @@ abstract void writeRaw(byte[] bytes, int offset, int length)
    * All data that has been written to the stream so far will be flushed.
    * New data can be still written to the stream while flushing is performed.
    */
-  abstract void setReadyToFlush() throws IOException;
+  abstract public void setReadyToFlush() throws IOException;
 
   /**
    * Flush and sync all data that is ready to be flush 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSEditLog.java

@@ -24,6 +24,7 @@
 import java.util.Collection;
 import java.util.Collections;
 import java.util.List;
+import java.lang.reflect.Constructor;
 
 import org.apache.commons.logging.Log;
 import org.apache.commons.logging.LogFactory;
@@ -31,6 +32,7 @@
 import org.apache.hadoop.classification.InterfaceStability;
 import org.apache.hadoop.fs.Options;
 import org.apache.hadoop.fs.permission.FsPermission;
+import org.apache.hadoop.hdfs.DFSConfigKeys;
 import org.apache.hadoop.hdfs.protocol.HdfsConstants;
 import org.apache.hadoop.hdfs.security.token.delegation.DelegationTokenIdentifier;
 import org.apache.hadoop.hdfs.server.common.HdfsServerConstants.NamenodeRole;
@@ -108,6 +110,7 @@ private enum State {
   private NameNodeMetrics metrics;
 
   private NNStorage storage;
+  private Configuration conf;
 
   private static class TransactionId {
     public long txid;
@@ -144,6 +147,7 @@ protected synchronized TransactionId initialValue() {
    * @param editsDirs List of journals to use
    */
   FSEditLog(Configuration conf, NNStorage storage, Collection<URI> editsDirs) {
+    this.conf = conf;
     isSyncRunning = false;
     this.storage = storage;
     metrics = NameNode.getNameNodeMetrics();
@@ -166,10 +170,14 @@ protected synchronized TransactionId initialValue() {
 
     this.journalSet = new JournalSet();
     for (URI u : this.editsDirs) {
+      if (u.getScheme().equals(NNStorage.LOCAL_URI_SCHEME)) {
         StorageDirectory sd = storage.getStorageDirectory(u);
         if (sd != null) {
           journalSet.add(new FileJournalManager(sd));
         }
+      } else {
+        journalSet.add(createJournal(u));
+      }
     }
  
     if (journalSet.isEmpty()) {
@@ -994,4 +1002,53 @@ static void closeAllStreams(Iterable<EditLogInputStream> streams) {
       IOUtils.closeStream(s);
     }
   }
+
+  /**
+   * Retrieve the implementation class for a Journal scheme.
+   * @param conf The configuration to retrieve the information from
+   * @param uriScheme The uri scheme to look up.
+   * @return the class of the journal implementation
+   * @throws IllegalArgumentException if no class is configured for uri
+   */
+  static Class<? extends JournalManager> getJournalClass(Configuration conf,
+                               String uriScheme) {
+    String key
+      = DFSConfigKeys.DFS_NAMENODE_EDITS_PLUGIN_PREFIX + "." + uriScheme;
+    Class <? extends JournalManager> clazz = null;
+    try {
+      clazz = conf.getClass(key, null, JournalManager.class);
+    } catch (RuntimeException re) {
+      throw new IllegalArgumentException(
+          "Invalid class specified for " + uriScheme, re);
+    }
+      
+    if (clazz == null) {
+      LOG.warn("No class configured for " +uriScheme
+               + ", " + key + " is empty");
+      throw new IllegalArgumentException(
+          "No class configured for " + uriScheme);
+    }
+    return clazz;
+  }
+
+  /**
+   * Construct a custom journal manager.
+   * The class to construct is taken from the configuration.
+   * @param uri Uri to construct
+   * @return The constructed journal manager
+   * @throws IllegalArgumentException if no class is configured for uri
+   */
+  private JournalManager createJournal(URI uri) {
+    Class<? extends JournalManager> clazz
+      = getJournalClass(conf, uri.getScheme());
+
+    try {
+      Constructor<? extends JournalManager> cons
+        = clazz.getConstructor(Configuration.class, URI.class);
+      return cons.newInstance(conf, uri);
+    } catch (Exception e) {
+      throw new IllegalArgumentException("Unable to construct journal, "
+                                         + uri, e);
+    }
+  }
 }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSEditLogLoader.java

@@ -25,6 +25,8 @@
 import java.util.Arrays;
 import java.util.EnumMap;
 
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
 import org.apache.hadoop.fs.permission.PermissionStatus;
 import org.apache.hadoop.hdfs.protocol.HdfsConstants;
 import org.apache.hadoop.hdfs.protocol.HdfsFileStatus;
@@ -57,6 +59,8 @@
 import org.apache.hadoop.hdfs.util.Holder;
 import com.google.common.base.Joiner;
 
+@InterfaceAudience.Private
+@InterfaceStability.Evolving
 public class FSEditLogLoader {
   private final FSNamesystem fsNamesys;
 
@@ -514,7 +518,7 @@ long getNumTransactions() {
   /**
    * Stream wrapper that keeps track of the current stream position.
    */
-  static class PositionTrackingInputStream extends FilterInputStream {
+  public static class PositionTrackingInputStream extends FilterInputStream {
     private long curPos = 0;
     private long markPos = -1;
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSEditLogOp.java

@@ -113,6 +113,10 @@ private FSEditLogOp(FSEditLogOpCodes opCode) {
     this.txid = 0;
   }
 
+  public long getTransactionId() {
+    return txid;
+  }
+
   public void setTransactionId(long txid) {
     this.txid = txid;
   }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/JournalManager.java

@@ -20,6 +20,8 @@
 import java.io.Closeable;
 import java.io.IOException;
 
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
 
 /**
  * A JournalManager is responsible for managing a single place of storing
@@ -28,7 +30,9 @@
  * each conceptual place of storage corresponds to exactly one instance of
  * this class, which is created when the EditLog is first opened.
  */
-interface JournalManager extends Closeable {
+@InterfaceAudience.Private
+@InterfaceStability.Evolving
+public interface JournalManager extends Closeable {
   /**
    * Begin writing to a new segment of the log stream, which starts at
    * the given transaction ID.
@@ -71,7 +75,6 @@ long getNumberOfTransactions(long fromTxnId)
    *
    * @param minTxIdToKeep the earliest txid that must be retained after purging
    *                      old logs
-   * @param purger the purging implementation to use
    * @throws IOException if purging fails
    */
   void purgeLogsOlderThan(long minTxIdToKeep)

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/JournalSet.java

@@ -309,7 +309,7 @@ private class JournalSetOutputStream extends EditLogOutputStream {
     }
 
     @Override
-    void write(final FSEditLogOp op)
+    public void write(final FSEditLogOp op)
         throws IOException {
       mapJournalsAndReportErrors(new JournalClosure() {
         @Override
@@ -322,7 +322,7 @@ public void apply(JournalAndStream jas) throws IOException {
     }
 
     @Override
-    void writeRaw(final byte[] data, final int offset, final int length)
+    public void writeRaw(final byte[] data, final int offset, final int length)
         throws IOException {
       mapJournalsAndReportErrors(new JournalClosure() {
         @Override
@@ -335,7 +335,7 @@ public void apply(JournalAndStream jas) throws IOException {
     }
 
     @Override
-    void create() throws IOException {
+    public void create() throws IOException {
       mapJournalsAndReportErrors(new JournalClosure() {
         @Override
         public void apply(JournalAndStream jas) throws IOException {
@@ -367,7 +367,7 @@ public void apply(JournalAndStream jas) throws IOException {
     }
 
     @Override
-    void setReadyToFlush() throws IOException {
+    public void setReadyToFlush() throws IOException {
       mapJournalsAndReportErrors(new JournalClosure() {
         @Override
         public void apply(JournalAndStream jas) throws IOException {

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/NNStorage.java

@@ -70,6 +70,7 @@ public class NNStorage extends Storage implements Closeable {
   private static final Log LOG = LogFactory.getLog(NNStorage.class.getName());
 
   static final String DEPRECATED_MESSAGE_DIGEST_PROPERTY = "imageMD5Digest";
+  static final String LOCAL_URI_SCHEME = "file";
 
   //
   // The filenames used for storing the images
@@ -324,22 +325,14 @@ StorageDirectory getStorageDirectory(URI uri) {
 
   /**
    * Checks the consistency of a URI, in particular if the scheme
-   * is specified and is supported by a concrete implementation
+   * is specified 
    * @param u URI whose consistency is being checked.
    */
   private static void checkSchemeConsistency(URI u) throws IOException {
     String scheme = u.getScheme();
     // the URI should have a proper scheme
-    if(scheme == null)
+    if(scheme == null) {
       throw new IOException("Undefined scheme for " + u);
-    else {
-      try {
-        // the scheme should be enumerated as JournalType
-        JournalType.valueOf(scheme.toUpperCase());
-      } catch (IllegalArgumentException iae){
-        throw new IOException("Unknown scheme " + scheme +
-            ". It should correspond to a JournalType enumeration value");
-      }
     }
   }
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/NameNodeResourceChecker.java

@@ -33,6 +33,8 @@
 import org.apache.hadoop.hdfs.server.common.Util;
 
 import com.google.common.annotations.VisibleForTesting;
+import com.google.common.collect.Collections2;
+import com.google.common.base.Predicate;
 
 /**
  * 
@@ -69,7 +71,18 @@ public NameNodeResourceChecker(Configuration conf) throws IOException {
         .getTrimmedStringCollection(DFSConfigKeys.DFS_NAMENODE_CHECKED_VOLUMES_KEY));
 
     addDirsToCheck(FSNamesystem.getNamespaceDirs(conf));
-    addDirsToCheck(FSNamesystem.getNamespaceEditsDirs(conf));
+    
+    Collection<URI> localEditDirs = Collections2.filter(
+        FSNamesystem.getNamespaceEditsDirs(conf),
+        new Predicate<URI>() {
+          public boolean apply(URI input) {
+            if (input.getScheme().equals(NNStorage.LOCAL_URI_SCHEME)) {
+              return true;
+            }
+            return false;
+          }
+        });
+    addDirsToCheck(localEditDirs);
     addDirsToCheck(extraCheckedVolumes);
   }