Hadoop 源码解读(一) - FSImage

FSImage 和Edits Log文件用于保存Namenode节点的元数据，用于持久化保存HDFS里各个数据文件之间的对应关系。FSImage在硬盘式以文件的方式保存集群中包括文件目录，数据块与相关datanode之间的映射关系。可能基于性能的考虑， FSImage并不是实时的更新以反映当前HDFS的文件及目录情况， 当前HDFS对于文件及目录等操作都以日志的形式保存于edits.log文件中，基于最小化停机时间的考虑，会存在一个备用的namenode节点， 通过IPC通信，定期的将edits.logs合并进FSImage中， 这样在HDFS下次重启时，namenode将花费较少的时间基于FSImage和edits.log文件在内存中重建HDFS。 感觉有点类似于oralce里面的dynamic check point. 

FSImage作为存储集群里面相关文件名及其一系列block与datanode的映射关系， 其存储结构又是怎么样呢？ 我们通过分析org.apache.hadoop.hdfs.server.namenode.FSImage可一窥究竟。

Hadoop1.2.1

boolean loadFSImage(File curFile) throws IOException {
    FSNamesystem fsNamesys = FSNamesystem.getFSNamesystem();
    FSDirectory fsDir = fsNamesys.dir;

    //
    // Load in bits
    //
    boolean needToSave = true;
    DataInputStream in = new DataInputStream(new BufferedInputStream(
                              new FileInputStream(curFile)));
    try {
      // read image version: first appeared in version -1
      //image版本号
      int imgVersion = in.readInt();
      // read namespaceID: first appeared in version -2
      //命名空间id
      this.namespaceID = in.readInt();

      // read number of files
      //文件或目录的数目，根据版本的不同，加以区别
      long numFiles;
      if (imgVersion <= -16) {
        numFiles = in.readLong();
      } else {
        numFiles = in.readInt();
      }

      this.layoutVersion = imgVersion;
      // read in the last generation stamp.
      //时间戳
      if (imgVersion <= -12) {
        long genstamp = in.readLong();
        fsNamesys.setGenerationStamp(genstamp); 
      }

      needToSave = (imgVersion != FSConstants.LAYOUT_VERSION);

      // read file info
      short replication = FSNamesystem.getFSNamesystem().getDefaultReplication();

      LOG.info("Number of files = " + numFiles);

      String path;
      String parentPath = "";
      INodeDirectory parentINode = fsDir.rootDir;
      //开始重建目录树
      for (long i = 0; i < numFiles; i++) {
        long modificationTime = 0;
        long atime = 0;
        long blockSize = 0;
        path = readString(in);//文件或者目录的路径名
        replication = in.readShort();//副本因子，默认为3，可配置 （如果是目录，这里应为0）
        replication = FSEditLog.adjustReplication(replication);
        modificationTime = in.readLong();//文件的mtime
        if (imgVersion <= -17) {
          atime = in.readLong();  //atime
        }
        if (imgVersion <= -8) {
          blockSize = in.readLong(); //block的大小，(目录为0)
        }
        int numBlocks = in.readInt(); //对应文件所包含的block总数，(目录为0)
        Block blocks[] = null;

        // for older versions, a blocklist of size 0
        // indicates a directory.
        if ((-9 <= imgVersion && numBlocks > 0) ||
            (imgVersion < -9 && numBlocks >= 0)) {
          blocks = new Block[numBlocks];
          for (int j = 0; j < numBlocks; j++) {
            blocks[j] = new Block();
            if (-14 < imgVersion) {
              blocks[j].set(in.readLong(), in.readLong(), 
                            Block.GRANDFATHER_GENERATION_STAMP);
            } else {
              blocks[j].readFields(in);
            }
          }
        }
        // Older versions of HDFS does not store the block size in inode.
        // If the file has more than one block, use the size of the 
        // first block as the blocksize. Otherwise use the default block size.
        //
        if (-8 <= imgVersion && blockSize == 0) {
          if (numBlocks > 1) {
            blockSize = blocks[0].getNumBytes();
          } else {
            long first = ((numBlocks == 1) ? blocks[0].getNumBytes(): 0);
            blockSize = Math.max(fsNamesys.getDefaultBlockSize(), first);
          }
        }
        
        // get quota only when the node is a directory
        long nsQuota = -1L;
        if (imgVersion <= -16 && blocks == null) {
          nsQuota = in.readLong();//nsQuota
        }
        long dsQuota = -1L;
        if (imgVersion <= -18 && blocks == null) {
          dsQuota = in.readLong();//dsQuota
        }
        
        PermissionStatus permissions = fsNamesys.getUpgradePermission();
        if (imgVersion <= -11) {
          permissions = PermissionStatus.read(in);
        }
        if (path.length() == 0) { // it is the root
          // update the root's attributes
          if (nsQuota != -1 || dsQuota != -1) {
            fsDir.rootDir.setQuota(nsQuota, dsQuota);
          }
          fsDir.rootDir.setModificationTime(modificationTime);
          fsDir.rootDir.setPermissionStatus(permissions);
          continue;
        }
        // check if the new inode belongs to the same parent
        if(!isParent(path, parentPath)) {
          parentINode = null;
          parentPath = getParent(path);
        }
        // add new inode
        parentINode = fsDir.addToParent(path, parentINode, permissions,
                                        blocks, replication, modificationTime, 
                                        atime, nsQuota, dsQuota, blockSize);
      }
      
      // load datanode info
      this.loadDatanodes(imgVersion, in);

      // load Files Under Construction
      this.loadFilesUnderConstruction(imgVersion, in, fsNamesys);
      
      this.loadSecretManagerState(imgVersion, in, fsNamesys);
      
    } finally {
      in.close();
    }
    
    return needToSave;
  }
  
  public void set(long blkid, long len, long genStamp) {
    this.blockId = blkid;
    this.numBytes = len;
    this.generationStamp = genStamp;
  }

 

imgVersion(int)：当前image的版本信息
namespaceID(int)：unknown
numFiles(long)：整个文件系统中包含有多少文件和目录
genStamp(long)：image的时间戳
path(String)：该目录或文件的路径，
replications(short)：副本数
mtime(long)：mtime
atime(long)：atime
blocksize(long)：目录的blocksize都为0
numBlocks(int)：实际有多少个文件块，目录的该值都为-1
if(numBlocks > 0){
  blockid(long)：属于该文件的block的blockid，
  numBytes(long)：该block的大小
  genStamp(long)：该block的时间戳
}
nsQuota(long)：namespace Quota值，若没加Quota限制则为-1
dsQuota(long)：disk Quota值，若没加限制则也为-1
...
..
.其他fields

Remark: 实际代码中对于不同版本的FSImage文件有一些分支判断

没有找到官方对于FSImage文件结构的描述,只能通过源码进行推断。  如果对于FSImage及edits.log文件结构清楚后， 应该可以实现脱离hadoop client 或者API对hdfs进行离线分析。 比如直接访问FSImage得到HDFS中的文件清单等信息，甚至直接定位到相关的datanode上的block