HDFS-4304. Make FSEditLogOp.MAX_OP_SIZE configurable. Contributed by Colin Patrick McCabe.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1449218 13f79535-47bb-0310-9956-ffa450edef68

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

@@ -309,9 +309,13 @@ Release 2.0.4-beta - UNRELEASED
   NEW FEATURES
 
   IMPROVEMENTS
+
     HDFS-4222. NN is unresponsive and loses heartbeats from DNs when 
     configured to use LDAP and LDAP has issues. (Xiaobo Peng, suresh)
 
+    HDFS-4304. Make FSEditLogOp.MAX_OP_SIZE configurable. (Colin Patrick
+    McCabe via atm)
+
   OPTIMIZATIONS
 
   BUG FIXES

hadoop-hdfs-project/hadoop-hdfs/src/contrib/bkjournal/src/main/java/org/apache/hadoop/contrib/bkjournal/BookKeeperEditLogInputStream.java

@@ -163,6 +163,11 @@ class BookKeeperEditLogInputStream extends EditLogInputStream {
     return ("BookKeeperEditLogInputStream {" + this.getName() + "}");
   }
 
+  @Override
+  public void setMaxOpSize(int maxOpSize) {
+    reader.setMaxOpSize(maxOpSize);
+  }
+
   /**
    * Input stream implementation which can be used by 
    * FSEditLogOp.Reader

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

@@ -392,6 +392,8 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
   public static final int     DFS_NAMENODE_CHECKED_VOLUMES_MINIMUM_DEFAULT = 1;
   public static final String  DFS_WEB_AUTHENTICATION_KERBEROS_PRINCIPAL_KEY = "dfs.web.authentication.kerberos.principal";
   public static final String  DFS_WEB_AUTHENTICATION_KERBEROS_KEYTAB_KEY = "dfs.web.authentication.kerberos.keytab";
+  public static final String  DFS_NAMENODE_MAX_OP_SIZE_KEY = "dfs.namenode.max.op.size";
+  public static final int     DFS_NAMENODE_MAX_OP_SIZE_DEFAULT = 50 * 1024 * 1024;
   
   public static final String DFS_BLOCK_LOCAL_PATH_ACCESS_USER_KEY = "dfs.block.local-path-access.user";
 

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

@@ -142,4 +142,9 @@ class EditLogBackupInputStream extends EditLogInputStream {
   public boolean isInProgress() {
     return true;
   }
+
+  @Override
+  public void setMaxOpSize(int maxOpSize) {
+    reader.setMaxOpSize(maxOpSize);
+  }
 }

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

@@ -32,6 +32,7 @@ import java.security.PrivilegedExceptionAction;
 import org.apache.commons.logging.Log;
 import org.apache.commons.logging.LogFactory;
 import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.hdfs.DFSConfigKeys;
 import org.apache.hadoop.hdfs.protocol.HdfsConstants;
 import org.apache.hadoop.hdfs.server.common.Storage;
 import org.apache.hadoop.hdfs.server.namenode.TransferFsImage.HttpGetFailedException;
@@ -53,6 +54,7 @@ public class EditLogFileInputStream extends EditLogInputStream {
   private final long firstTxId;
   private final long lastTxId;
   private final boolean isInProgress;
+  private int maxOpSize;
   static private enum State {
     UNINIT,
     OPEN,
@@ -118,6 +120,7 @@ public class EditLogFileInputStream extends EditLogInputStream {
     this.firstTxId = firstTxId;
     this.lastTxId = lastTxId;
     this.isInProgress = isInProgress;
+    this.maxOpSize = DFSConfigKeys.DFS_NAMENODE_MAX_OP_SIZE_DEFAULT;
   }
 
   private void init() throws LogHeaderCorruptException, IOException {
@@ -134,6 +137,7 @@ public class EditLogFileInputStream extends EditLogInputStream {
         throw new LogHeaderCorruptException("No header found in log");
       }
       reader = new FSEditLogOp.Reader(dataIn, tracker, logVersion);
+      reader.setMaxOpSize(maxOpSize);
       state = State.OPEN;
     } finally {
       if (reader == null) {
@@ -413,4 +417,11 @@ public class EditLogFileInputStream extends EditLogInputStream {
     }
   }
 
+  @Override
+  public void setMaxOpSize(int maxOpSize) {
+    this.maxOpSize = maxOpSize;
+    if (reader != null) {
+      reader.setMaxOpSize(maxOpSize);
+    }
+  }
 }

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

@@ -165,4 +165,9 @@ public abstract class EditLogInputStream implements Closeable {
    * Return true if this stream is in progress, false if it is finalized.
    */
   public abstract boolean isInProgress();
+  
+  /**
+   * Set the maximum opcode size in bytes.
+   */
+  public abstract void setMaxOpSize(int maxOpSize);
 }

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

@@ -51,6 +51,7 @@ import org.apache.hadoop.io.WritableFactory;
 import org.apache.hadoop.hdfs.util.XMLUtils;
 import org.apache.hadoop.hdfs.util.XMLUtils.InvalidXmlException;
 import org.apache.hadoop.hdfs.util.XMLUtils.Stanza;
+import org.apache.hadoop.hdfs.DFSConfigKeys;
 import org.apache.hadoop.hdfs.DeprecatedUTF8;
 import org.xml.sax.ContentHandler;
 import org.xml.sax.SAXException;
@@ -75,11 +76,6 @@ import java.io.EOFException;
 public abstract class FSEditLogOp {
   public final FSEditLogOpCodes opCode;
   long txid;
-  /**
-   * Opcode size is limited to 1.5 megabytes
-   */
-  public static final int MAX_OP_SIZE = (3 * 1024 * 1024) / 2;
-
 
   @SuppressWarnings("deprecation")
   final public static class OpInstanceCache {
@@ -2246,6 +2242,7 @@ public abstract class FSEditLogOp {
     private final int logVersion;
     private final Checksum checksum;
     private final OpInstanceCache cache;
+    private int maxOpSize;
 
     /**
      * Construct the reader
@@ -2253,7 +2250,8 @@ public abstract class FSEditLogOp {
      * @param logVersion The version of the data coming from the stream.
      */
     @SuppressWarnings("deprecation")
-    public Reader(DataInputStream in, StreamLimiter limiter, int logVersion) {
+    public Reader(DataInputStream in, StreamLimiter limiter,
+        int logVersion) {
       this.logVersion = logVersion;
       if (LayoutVersion.supports(Feature.EDITS_CHESKUM, logVersion)) {
         this.checksum = new PureJavaCrc32();
@@ -2269,6 +2267,11 @@ public abstract class FSEditLogOp {
       }
       this.limiter = limiter;
       this.cache = new OpInstanceCache();
+      this.maxOpSize = DFSConfigKeys.DFS_NAMENODE_MAX_OP_SIZE_DEFAULT;
+    }
+
+    public void setMaxOpSize(int maxOpSize) {
+      this.maxOpSize = maxOpSize;
     }
 
     /**
@@ -2363,8 +2366,8 @@ public abstract class FSEditLogOp {
      * problematic byte.  This usually means the beginning of the opcode.
      */
     private FSEditLogOp decodeOp() throws IOException {
-      limiter.setLimit(MAX_OP_SIZE);
+      limiter.setLimit(maxOpSize);
-      in.mark(MAX_OP_SIZE);
+      in.mark(maxOpSize);
 
       if (checksum != null) {
         checksum.reset();

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

@@ -607,6 +607,12 @@ public class FSImage implements Closeable {
       editStreams = FSImagePreTransactionalStorageInspector
         .getEditLogStreams(storage);
     }
+    int maxOpSize = conf.getInt(DFSConfigKeys.
+          DFS_NAMENODE_MAX_OP_SIZE_KEY,
+        DFSConfigKeys.DFS_NAMENODE_MAX_OP_SIZE_DEFAULT);
+    for (EditLogInputStream elis : editStreams) {
+      elis.setMaxOpSize(maxOpSize);
+    }
  
     LOG.debug("Planning to load image :\n" + imageFile);
     for (EditLogInputStream l : editStreams) {

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

@@ -267,4 +267,11 @@ class RedundantEditLogInputStream extends EditLogInputStream {
       super(msg);
     }
   }
+
+  @Override
+  public void setMaxOpSize(int maxOpSize) {
+    for (EditLogInputStream elis : streams) {
+      elis.setMaxOpSize(maxOpSize);
+    }
+  }
 }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestEditLog.java

@@ -861,6 +861,11 @@ public class TestEditLog {
     public boolean isInProgress() {
       return true;
     }
+
+    @Override
+    public void setMaxOpSize(int maxOpSize) {
+      reader.setMaxOpSize(maxOpSize);
+    }
   }
 
   @Test

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestNameNodeRecovery.java

@@ -83,6 +83,7 @@ public class TestNameNodeRecovery {
       elfos.close();
       elfos = null;
       elfis = new EditLogFileInputStream(TEST_LOG_NAME);
+      elfis.setMaxOpSize(elts.getMaxOpSize());
       
       // reading through normally will get you an exception
       Set<Long> validTxIds = elts.getValidTxIds();
@@ -143,7 +144,7 @@ public class TestNameNodeRecovery {
   /**
    * A test scenario for the edit log
    */
-  private interface EditLogTestSetup {
+  private static abstract class EditLogTestSetup {
     /** 
      * Set up the edit log.
      */
@@ -162,6 +163,13 @@ public class TestNameNodeRecovery {
      * edit log.
      **/
     abstract public Set<Long> getValidTxIds();
+    
+    /**
+     * Return the maximum opcode size we will use for input.
+     */
+    public int getMaxOpSize() {
+      return DFSConfigKeys.DFS_NAMENODE_MAX_OP_SIZE_DEFAULT;
+    }
   }
   
   static void padEditLog(EditLogOutputStream elos, int paddingLength)
@@ -182,10 +190,10 @@ public class TestNameNodeRecovery {
   }
 
   static void addDeleteOpcode(EditLogOutputStream elos,
-        OpInstanceCache cache) throws IOException {
+        OpInstanceCache cache, long txId, String path) throws IOException {
     DeleteOp op = DeleteOp.getInstance(cache);
-    op.setTransactionId(0x0);
+    op.setTransactionId(txId);
-    op.setPath("/foo");
+    op.setPath(path);
     op.setTimestamp(0);
     elos.write(op);
   }
@@ -198,7 +206,7 @@ public class TestNameNodeRecovery {
    * able to handle any amount of padding (including no padding) without
    * throwing an exception.
    */
-  private static class EltsTestEmptyLog implements EditLogTestSetup {
+  private static class EltsTestEmptyLog extends EditLogTestSetup {
     private int paddingLength;
 
     public EltsTestEmptyLog(int paddingLength) {
@@ -242,6 +250,42 @@ public class TestNameNodeRecovery {
         3 * EditLogFileOutputStream.MIN_PREALLOCATION_LENGTH));
   }
 
+  /**
+   * Test using a non-default maximum opcode length.
+   */
+  private static class EltsTestNonDefaultMaxOpSize extends EditLogTestSetup {
+    public EltsTestNonDefaultMaxOpSize() {
+    }
+
+    @Override
+    public void addTransactionsToLog(EditLogOutputStream elos,
+        OpInstanceCache cache) throws IOException {
+      addDeleteOpcode(elos, cache, 0, "/foo");
+      addDeleteOpcode(elos, cache, 1,
+       "/supercalifragalisticexpialadocius.supercalifragalisticexpialadocius");
+    }
+
+    @Override
+    public long getLastValidTxId() {
+      return 0;
+    }
+
+    @Override
+    public Set<Long> getValidTxIds() {
+      return Sets.newHashSet(0L);
+    } 
+    
+    public int getMaxOpSize() {
+      return 30;
+    }
+  }
+
+  /** Test an empty edit log with extra-long padding */
+  @Test(timeout=180000)
+  public void testNonDefaultMaxOpSize() throws IOException {
+    runEditLogTest(new EltsTestNonDefaultMaxOpSize());
+  }
+
   /**
    * Test the scenario where an edit log contains some padding (0xff) bytes
    * followed by valid opcode data.
@@ -249,7 +293,7 @@ public class TestNameNodeRecovery {
    * These edit logs are corrupt, but all the opcodes should be recoverable
    * with recovery mode.
    */
-  private static class EltsTestOpcodesAfterPadding implements EditLogTestSetup {
+  private static class EltsTestOpcodesAfterPadding extends EditLogTestSetup {
     private int paddingLength;
 
     public EltsTestOpcodesAfterPadding(int paddingLength) {
@@ -260,7 +304,7 @@ public class TestNameNodeRecovery {
     public void addTransactionsToLog(EditLogOutputStream elos,
         OpInstanceCache cache) throws IOException {
       padEditLog(elos, paddingLength);
-      addDeleteOpcode(elos, cache);
+      addDeleteOpcode(elos, cache, 0, "/foo");
     }
 
     @Override
@@ -286,7 +330,7 @@ public class TestNameNodeRecovery {
         3 * EditLogFileOutputStream.MIN_PREALLOCATION_LENGTH));
   }
 
-  private static class EltsTestGarbageInEditLog implements EditLogTestSetup {
+  private static class EltsTestGarbageInEditLog extends EditLogTestSetup {
     final private long BAD_TXID = 4;
     final private long MAX_TXID = 10;