hdfs-<NameNode的format>
1、在属性dfs.namenode.name.dir指定的目录中创建相应的元数据文件

a)current/VERSION

#Fri Dec 06 11:21:35 CST 2013namespaceID=140437801clusterID=CID-5a02557f-1977-44d6-b7bc-d124be2d0ba3cTime=0storageType=NAME_NODEblockpoolID=BP-599437854-10.12.120.79-1386299958223layoutVersion=-47

b)current/seen_txid
写入事务序号txid

0

FSImage$FSImageSaver写入current/fsimage.ckpt
然后再重命名为fsimage(+19位的txid)fsimage_0000000000000000000
1、文件头

out.writeInt(HdfsConstants.LAYOUT_VERSION);out.writeInt(sourceNamesystem.unprotectedGetNamespaceInfo()            .getNamespaceID());out.writeLong(fsDir.rootDir.numItemsInTree());//整个文件系统中包含有多少文件和目录out.writeLong(sourceNamesystem.getGenerationStampV1());out.writeLong(sourceNamesystem.getGenerationStampV2());       out.writeLong(sourceNamesystem.getGenerationStampAtblockIdSwitch());out.writeLong(sourceNamesystem.getLastAllocatedBlockId());out.writeLong(context.getTxId());out.writeLong(sourceNamesystem.getLastInodeId());

2、接下来便是目录和文件的元数据信息

Map<String, INodeFileUnderConstruction> nodes =          leaseManager.getINodesUnderConstruction();      out.writeInt(nodes.size()); // write the size          for (Map.Entry<String, INodeFileUnderConstruction> entry           : nodes.entrySet()) {        FSImageSerialization.writeINodeUnderConstruction(            out, entry.getValue(), entry.getKey());
writeString(path, out);    out.writeLong(cons.getId());    out.writeShort(cons.getFileReplication());    out.writeLong(cons.getModificationTime());    out.writeLong(cons.getPreferredBlockSize());    writeBlocks(cons.getBlocks(), out);    cons.getPermissionStatus().write(out);    writeString(cons.getClientName(), out);    writeString(cons.getClientMachine(), out);    out.writeInt(0); //  do not store locations of last block

      }

参考：http://abloz.com/2013/01/08/hadoop-1-0-4-fsimage-file-format.html
http://blog.csdn.net/xhh198781/article/details/6904615