Hadoop学习十二：Hadoop-Hdfs Storage源码

一.物理文件夹和文件与类的对应关系

•  Storage：存储信息文件的集合。由一系列StorageDirectory组成，这些StorageDirectory的VERSION相同。Storage对应着Hdfs的dfs.data.dir或dfs.name.dir目录，默认为%hadoop_home%/dfs/data或%hadoop_home%/dfs/name。DataNode可以配置多个dfs.data.dir，如dfs/data,dfs/data2，此时就有两个Sorage。
• StorageDirectory：一个存储信息文件。代表Storage下涉及到升级方面的文件，它们共享一个in_use.lock(图上未标出这层关系)。
• StorageInfo：存储信息文件的父类。

二.类图

 

三.StorageInfo

public class StorageInfo {
	//Hadoop版本号，如果Hadoop调整文件结构布局，版本号就会修改，这样可以保证文件结构和应用一致
	//参考FSConstants.LAYOUT_VERSION=-32的定义：
	// Version is reflected in the dfs image and edit log files.		Version代表image和editlos存储文件的结构
	// Version is reflected in the data storage file.		Version代表data存储文件的结构
	// Versions are negative.		Versions是负数。
	// Decrement LAYOUT_VERSION to define a new version.
	public int   layoutVersion;  
	//NameNode节点的namespaceID在它format是生成，每格式化一次，就会产生一个新的namespaceID
	//DataNode节点的每一个StorageDirectory的namespaceID必须与NameNode节点的namespaceID保持一致
	public int   namespaceID;  
	//FsImage format时赋值
	public long  cTime;          // creation timestamp
}

//参考FsImage.format()
//public void format() throws IOException {
//    this.layoutVersion = FSConstants.LAYOUT_VERSION;
//    this.namespaceID = newNamespaceID();
//    this.cTime = 0L;
//    this.checkpointTime = FSNamesystem.now();
//	   ...
//    }
//  }

 

四.StorageDirectory

  //一个Storage包含一系列StorageDirectory，它们共享一个in_use.lock
  public class StorageDirectory {
    File              root; 		//对应dfs.data.dir或dfs.name.dir目录
    FileLock          lock;	//对应dfs.data.dir或dfs.name.dir目录下in_use.lock
    StorageDirType dirType; // storage dir type
    
    
    //读取dfs.data.dir/current/VERSION或dfs.name.dir/current/VERSION
    //read里调用getFields，读取dfs.data.dir/current/VERSION或dfs.name.dir/current/VERSION到this
    public void read() throws IOException {
      read(getVersionFile());
    }
    
    //write里调用setFields，把this的属性写到dfs.data.dir/current/VERSION或dfs.name.dir/current/VERSION
    public void write() throws IOException {
      corruptPreUpgradeStorage(root);
      write(getVersionFile());
    }

    //读取Storage下任何一个StorageDirectory
    public File get*Dir() {
      return new File(root,  ...);
    }


    //During startup Hadoop servers (name-node and data-nodes) read their local 
 * storage information from them.
    //根据启动参数判断系统处于哪个状态
    public StorageState analyzeStorage(StartupOption startOpt) throws IOException {
    	//见图状态分析
    }

    //Complete or recover storage state from previously failed transition.
    public void doRecover(StorageState curState) throws IOException {
    	//见图恢复操作
    }

    //lock this storage 
    public void lock() throws IOException {
      this.lock = tryLock();
    }

    //unlock this storage
    public void unlock() throws IOException {
      this.lock.release();
      lock.channel().close();
      lock = null;
    }
  }

 

五.Storage

public abstract class Storage extends StorageInfo {
  
  private   static final String STORAGE_FILE_LOCK     = "in_use.lock";
  protected static final String STORAGE_FILE_VERSION  = "VERSION";
  public static final String STORAGE_DIR_CURRENT   = "current";
  private   static final String STORAGE_DIR_PREVIOUS  = "previous";
  private   static final String STORAGE_TMP_REMOVED   = "removed.tmp";
  private   static final String STORAGE_TMP_PREVIOUS  = "previous.tmp";
  private   static final String STORAGE_TMP_FINALIZED = "finalized.tmp";
  private   static final String STORAGE_TMP_LAST_CKPT = "lastcheckpoint.tmp";
  private   static final String STORAGE_PREVIOUS_CKPT = "previous.checkpoint";
  
  public enum StorageState {
    NON_EXISTENT,
    NOT_FORMATTED,
    COMPLETE_UPGRADE,
    RECOVER_UPGRADE,
    COMPLETE_FINALIZE,
    COMPLETE_ROLLBACK,
    RECOVER_ROLLBACK,
    COMPLETE_CHECKPOINT,
    RECOVER_CHECKPOINT,
    NORMAL;
  }
  
  //StorageDirectory下的文件类型
  public interface StorageDirType {
    public StorageDirType getStorageDirType();
    public boolean isOfType(StorageDirType type);
  }
  
  //节点类型
  private NodeType storageType;    // Type of the node using this storage 
  protected List<StorageDirectory> storageDirs = new ArrayList<StorageDirectory>();
  
  //迭代Storage包含的所有StorageDirectory
  public Iterator<StorageDirectory> dirIterator(StorageDirType dirType) {
    return new DirIterator(dirType);
  }
  
  protected Storage(NodeType type) {
    super();
    this.storageType = type;
  }
  
  //StorageDirectory.read()调用此方法，读取dfs.data.dir/current/VERSION或dfs.name.dir/current/VERSION到this
  //因为所有的StorageDirectory具有相同的VERSION，所以可以把任意一个StorageDirectory的VERSION写到this的这四个属性
  //write同read
  protected void getFields(Properties props,  StorageDirectory sd ) throws IOException {
    String sv, st, sid, sct;
    sv = props.getProperty("layoutVersion");
    st = props.getProperty("storageType");
    sid = props.getProperty("namespaceID");
    sct = props.getProperty("cTime");
    layoutVersion = rv;
    storageType = rt;
    namespaceID = rid;
    cTime = rct;
  }
  
  //StorageDirectory.write()调用此方法，把this的属性写到dfs.data.dir/current/VERSION或dfs.name.dir/current/VERSION
  protected void setFields(Properties props, StorageDirectory sd ) throws IOException {
    props.setProperty("layoutVersion", String.valueOf(layoutVersion));
    props.setProperty("storageType", storageType.toString());
    props.setProperty("namespaceID", String.valueOf(namespaceID));
    props.setProperty("cTime", String.valueOf(cTime));
  }

}

 

 

六.VERSION例子

 

#Sun May 12 10:25:01 CST 2013
namespaceID=1378739863
storageID=DS-1718846927-192.168.1.164-50010-1368305080745
cTime=0
storageType=DATA_NODE
layoutVersion=-32

#Mon May 13 03:26:48 CST 2013
namespaceID=1378739863
cTime=0
storageType=NAME_NODE
layoutVersion=-32

 

 

七.最核心的方法StorageDirectory.analyzeStorage和doRecover

• 这两个方法就是对整个Storage状态分析和恢复操作。
• 下文会结合DataNode启动流程把整个串起来。本文先了解这两个方法，当然你必须先了解Hadoop的系统状态。
• 说明：第二张图少了一步。lastcheckpoint.tmp存在之前，首先分析当前StorageDirectory的root是否存在或是否可写或是否文件夹，不满足以上任何一点返回状态StorageState.NON_EXISTENT；满足后才是判断lastcheckpoint.tmp存在。

八.感谢作者的最后一张图

http://www.cnblogs.com/xuxm2007/archive/2012/07/04/2576745.html

评论

1 楼 chen030112 2014-04-13  

很喜欢博主的文章，刚刚用豆约翰博客备份专家备份了您的全部博文。