You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-commits@hadoop.apache.org by ki...@apache.org on 2013/11/25 16:40:31 UTC

svn commit: r1545326 - in /hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs: CHANGES.txt src/main/java/org/apache/hadoop/hdfs/server/datanode/DataStorage.java src/test/java/org/apache/hadoop/hdfs/TestDFSRollback.java

Author: kihwal
Date: Mon Nov 25 15:40:31 2013
New Revision: 1545326

URL: http://svn.apache.org/r1545326
Log:
svn merge -c 1545322 merging from trunk to branch-0.23 to fix HDFS-5266. Datanode cannot roll back to previous layout version.

Modified:
    hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt
    hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DataStorage.java
    hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDFSRollback.java

Modified: hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt
URL: http://svn.apache.org/viewvc/hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt?rev=1545326&r1=1545325&r2=1545326&view=diff
==============================================================================
--- hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt (original)
+++ hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt Mon Nov 25 15:40:31 2013
@@ -60,6 +60,8 @@ Release 0.23.10 - UNRELEASED
 
     HDFS-5438. Flaws in block report processing can cause data loss. (kihwal)
 
+    HDFS-5526. Datanode cannot roll back to previous layout version (kihwal)
+
 Release 0.23.9 - 2013-07-08
 
   INCOMPATIBLE CHANGES

Modified: hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DataStorage.java
URL: http://svn.apache.org/viewvc/hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DataStorage.java?rev=1545326&r1=1545325&r2=1545326&view=diff
==============================================================================
--- hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DataStorage.java (original)
+++ hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DataStorage.java Mon Nov 25 15:40:31 2013
@@ -303,7 +303,16 @@ public class DataStorage extends Storage
   @Override
   protected void setFieldsFromProperties(Properties props, StorageDirectory sd)
       throws IOException {
-    setLayoutVersion(props, sd);
+    setFieldsFromProperties(props, sd, false, 0);
+  }
+
+  private void setFieldsFromProperties(Properties props, StorageDirectory sd,
+      boolean overrideLayoutVersion, int toLayoutVersion) throws IOException {
+    if (overrideLayoutVersion) {
+      this.layoutVersion = toLayoutVersion;
+    } else {
+      setLayoutVersion(props, sd);
+    }
     setcTime(props, sd);
     setStorageType(props, sd);
     setClusterId(props, layoutVersion, sd);
@@ -351,13 +360,20 @@ public class DataStorage extends Storage
     return true;
   }
   
+  /** Read VERSION file for rollback */
+  void readProperties(StorageDirectory sd, int rollbackLayoutVersion)
+      throws IOException {
+    Properties props = readPropertiesFile(sd.getVersionFile());
+    setFieldsFromProperties(props, sd, true, rollbackLayoutVersion);
+  }
+
   /**
    * Analize which and whether a transition of the fs state is required
    * and perform it if necessary.
    * 
-   * Rollback if previousLV >= LAYOUT_VERSION && prevCTime <= namenode.cTime
-   * Upgrade if this.LV > LAYOUT_VERSION || this.cTime < namenode.cTime
-   * Regular startup if this.LV = LAYOUT_VERSION && this.cTime = namenode.cTime
+   * Rollback if the rollback startup option was specified.
+   * Upgrade if this.LV > LAYOUT_VERSION
+   * Regular startup if this.LV = LAYOUT_VERSION
    * 
    * @param datanode Datanode to which this storage belongs to
    * @param sd  storage directory
@@ -397,9 +413,11 @@ public class DataStorage extends Storage
           + nsInfo.getClusterID() + "; datanode clusterID = " + getClusterID());
     }
     
-    // regular start up
-    if (this.layoutVersion == HdfsConstants.LAYOUT_VERSION 
-        && this.cTime == nsInfo.getCTime())
+    // After addition of the federation feature, ctime check is only 
+    // meaningful at BlockPoolSliceStorage level. 
+
+    // regular start up. 
+    if (this.layoutVersion == HdfsConstants.LAYOUT_VERSION)
       return; // regular startup
     // verify necessity of a distributed upgrade
     UpgradeManagerDatanode um = 
@@ -407,19 +425,20 @@ public class DataStorage extends Storage
     verifyDistributedUpgradeProgress(um, nsInfo);
     
     // do upgrade
-    if (this.layoutVersion > HdfsConstants.LAYOUT_VERSION
-        || this.cTime < nsInfo.getCTime()) {
+    if (this.layoutVersion > HdfsConstants.LAYOUT_VERSION) {
       doUpgrade(sd, nsInfo);  // upgrade
       return;
     }
     
-    // layoutVersion == LAYOUT_VERSION && this.cTime > nsInfo.cTime
-    // must shutdown
-    throw new IOException("Datanode state: LV = " + this.getLayoutVersion() 
-                          + " CTime = " + this.getCTime() 
-                          + " is newer than the namespace state: LV = "
-                          + nsInfo.getLayoutVersion() 
-                          + " CTime = " + nsInfo.getCTime());
+    // layoutVersion < LAYOUT_VERSION. I.e. stored layout version is newer
+    // than the version supported by datanode. This should have been caught
+    // in readProperties(), even if rollback was not carried out or somehow
+    // failed.
+    throw new IOException("BUG: The stored LV = " + this.getLayoutVersion()
+                          + " is newer than the supported LV = "
+                          + HdfsConstants.LAYOUT_VERSION
+                          + " or name node LV = "
+                          + nsInfo.getLayoutVersion());
   }
 
   /**
@@ -445,8 +464,13 @@ public class DataStorage extends Storage
    * @throws IOException on error
    */
   void doUpgrade(StorageDirectory sd, NamespaceInfo nsInfo) throws IOException {
+    // If the existing on-disk layout version supportes federation, simply
+    // update its layout version.
     if (LayoutVersion.supports(Feature.FEDERATION, layoutVersion)) {
-      clusterID = nsInfo.getClusterID();
+      // The VERSION file is already read in. Override the layoutVersion 
+      // field and overwrite the file.
+      LOG.info("Updating layout version from " + layoutVersion + " to "
+          + nsInfo.getLayoutVersion() + " for storage " + sd.getRoot());
       layoutVersion = nsInfo.getLayoutVersion();
       writeProperties(sd);
       return;
@@ -531,15 +555,32 @@ public class DataStorage extends Storage
    * <li> Remove removed.tmp </li>
    * </ol>
    * 
-   * Do nothing, if previous directory does not exist.
+   * If previous directory does not exist and the current version supports
+   * federation, perform a simple rollback of layout version. This does not
+   * involve saving/restoration of actual data.
    */
   void doRollback( StorageDirectory sd,
                    NamespaceInfo nsInfo
                    ) throws IOException {
     File prevDir = sd.getPreviousDir();
-    // regular startup if previous dir does not exist
-    if (!prevDir.exists())
+    // This is a regular startup or a post-federation rollback
+    if (!prevDir.exists()) {
+      // The current datanode version supports federation and the layout
+      // version from namenode matches what the datanode supports. An invalid
+      // rollback may happen if namenode didn't rollback and datanode is
+      // running a wrong version.  But this will be detected in block pool
+      // level and the invalid VERSION content will be overwritten when
+      // the error is corrected and rollback is retried.
+      if (LayoutVersion.supports(Feature.FEDERATION,
+          HdfsConstants.LAYOUT_VERSION) && 
+          HdfsConstants.LAYOUT_VERSION == nsInfo.getLayoutVersion()) {
+        readProperties(sd, nsInfo.getLayoutVersion());
+        writeProperties(sd);
+        LOG.info("Layout version rolled back to " +
+            nsInfo.getLayoutVersion() + " for storage " + sd.getRoot());
+      }
       return;
+    }
     DataStorage prevInfo = new DataStorage();
     prevInfo.readPreviousVersionProperties(sd);
 

Modified: hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDFSRollback.java
URL: http://svn.apache.org/viewvc/hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDFSRollback.java?rev=1545326&r1=1545325&r2=1545326&view=diff
==============================================================================
--- hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDFSRollback.java (original)
+++ hadoop/common/branches/branch-0.23/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDFSRollback.java Mon Nov 25 15:40:31 2013
@@ -190,21 +190,25 @@ public class TestDFSRollback extends Tes
       // Create a previous snapshot for the blockpool
       UpgradeUtilities.createBlockPoolStorageDirs(dataNodeDirs, "previous",
           UpgradeUtilities.getCurrentBlockPoolID(cluster));
-      // Older LayoutVersion to make it rollback
+      // Put newer layout version in current.
       storageInfo = new StorageInfo(
-          UpgradeUtilities.getCurrentLayoutVersion()+1,
+          UpgradeUtilities.getCurrentLayoutVersion()-1,
           UpgradeUtilities.getCurrentNamespaceID(cluster),
           UpgradeUtilities.getCurrentClusterID(cluster),
           UpgradeUtilities.getCurrentFsscTime(cluster));
-      // Create old VERSION file for each data dir
+
+      // Overwrite VERSION file in the current directory of
+      // volume directories and block pool slice directories
+      // with a layout version from future.
+      File[] dataCurrentDirs = new File[dataNodeDirs.length];
       for (int i=0; i<dataNodeDirs.length; i++) {
-        Path bpPrevPath = new Path(dataNodeDirs[i] + "/current/"
-            + UpgradeUtilities.getCurrentBlockPoolID(cluster));
-        UpgradeUtilities.createBlockPoolVersionFile(
-            new File(bpPrevPath.toString()),
-            storageInfo,
-            UpgradeUtilities.getCurrentBlockPoolID(cluster));
+        dataCurrentDirs[i] = new File((new Path(dataNodeDirs[i] 
+            + "/current")).toString());
       }
+      UpgradeUtilities.createDataNodeVersionFile(
+          dataCurrentDirs,
+          storageInfo,
+          UpgradeUtilities.getCurrentBlockPoolID(cluster));
 
       cluster.startDataNodes(conf, 1, false, StartupOption.ROLLBACK, null);
       assertTrue(cluster.isDataNodeUp());