hive创建目录时相关的几个hdfs中的类:
org.apache.hadoop.hdfs.DistributedFileSystem,FileSystem 的具体实现类 org.apache.hadoop.hdfs.DFSClient,client操作hdfs文件系统的类 org.apache.hadoop.fs.permission.FsPermission 文件权限相关类,主要的方法有getUMask和applyUMask方法
org.apache.hadoop.hdfs.DistributedFileSystem中需要注意的几个方法:
initialize,主要用来初始DFSClient的实例:
@Override public void initialize(URI uri, Configuration conf) throws IOException { super.initialize(uri, conf); setConf(conf); String host = uri.getHost(); if (host == null) { throw new IOException("Incomplete HDFS URI, no host: "+ uri); } this.dfs = new DFSClient(uri, conf, statistics); this.uri = URI.create(uri.getScheme()+"://"+uri.getAuthority()); this.workingDir = getHomeDirectory(); }
mkdir用来创建一个目录,mkdirs用来创建多个目录(类似于mkdir -p):
public boolean mkdir(Path f, FsPermission permission) throws IOException { statistics.incrementWriteOps(1); return dfs.mkdirs(getPathName(f), permission, false); } public boolean mkdirs(Path f, FsPermission permission) throws IOException { statistics.incrementWriteOps(1); return dfs.mkdirs(getPathName(f), permission, true); }
两者最终调用的都是DFSClient.mkdirs方法,org.apache.hadoop.hdfs.DFSClient的mkdirs方法:
final Conf dfsClientConf; ... public boolean mkdirs(String src, FsPermission permission, boolean createParent) throws IOException { if (permission == null) { //如果传入的权限为null permission = FsPermission.getDefault(); } FsPermission masked = permission.applyUMask(dfsClientConf.uMask); return primitiveMkdir(src, masked, createParent); //调用primitiveMkdir方法 }
这里需要注意 FsPermission.getDefault方法和Conf.uMask属性(Conf是DFSClient的内部类,主要用来设置默认配置)
Conf.uMask属性:
uMask = FsPermission.getUMask(conf); //由getUMask获取
getUMask方法:
public static final String DEPRECATED_UMASK_LABEL = "dfs.umask"; public static final String UMASK_LABEL = CommonConfigurationKeys.FS_PERMISSIONS_UMASK_KEY; //fs.permissions.umask-mode public static final int DEFAULT_UMASK = CommonConfigurationKeys.FS_PERMISSIONS_UMASK_DEFAULT; //0022 public static FsPermission getUMask(Configuration conf) { int umask = DEFAULT_UMASK; if(conf != null) { String confUmask = conf.get(UMASK_LABEL); int oldUmask = conf.getInt(DEPRECATED_UMASK_LABEL, Integer.MIN_VALUE); //老的配置项:dfs.umask,默认值为Integer.MIN_VALUE(-2147483648) try { if(confUmask != null) { //如果设置了fs.permissions.umask-mode,则按这个umask,否则为默认的umask(0022) umask = new UmaskParser(confUmask).getUMask(); } } catch(IllegalArgumentException iae) { // Provide more explanation for user-facing message String type = iae instanceof NumberFormatException ? "decimal" : "octal or symbolic"; String error = "Unable to parse configuration " + UMASK_LABEL + " with value " + confUmask + " as " + type + " umask."; LOG.warn(error); // If oldUmask is not set, then throw the exception if (oldUmask == Integer.MIN_VALUE) { throw new IllegalArgumentException(error); } } if(oldUmask != Integer.MIN_VALUE) { //如果手动设置了老的配置项dfs.umask if (umask != oldUmask) { //并且dfs.umask的值不等于0022 LOG.warn(DEPRECATED_UMASK_LABEL + " configuration key is deprecated. " + "Convert to " + UMASK_LABEL + ", using octal or symbolic umask " + "specifications."); // Old and new umask values do not match - Use old umask umask = oldUmask; //umask为默认值0022 } } } return new FsPermission((short)umask); }
在hive中创建hdfs的目录有两种方法
1)通过Utilities的createDirsWithPermission方法,这种方法会重设fs.permissions.umask-mode
2)直接通过DistributedFileSystem的mkdirs方法创建
两者最终都是调用了DFSClient的mkdirs方法,不同的是调用Utilities.createDirsWithPermission创建的目录权限在proxy时权限有可能是777(因为手动设置了权限为777),
比如:
Context类的构造函数中创建临时文件目录通过Context.getMRScratchDir调getLocalScratchDir(local job)或getScratchDir(非local job),其中getScratchDir中调用Utilities.createDirsWithPermission方法调用目录
public static boolean createDirsWithPermission(Configuration conf, Path mkdirPath, FsPermission fsPermission, boolean recursive) throws IOException { String origUmask = null; LOG.warn("Create dirs " + mkdirPath + " with permission " + fsPermission + " recursive " + recursive); if (recursive) { //如果recursive为true,设置fs.permissions.umask-mode为000, //默认情况下recursive = SessionState.get().isHiveServerQuery() &&conf.getBoolean(HiveConf.ConfVars.HIVE_SERVER2_ENABLE_DOAS.varname,HiveConf.ConfVars.HIVE_SERVER2_ENABLE_DOAS.defaultBoolVal); //即时来自hiveserver的请求,并且开启了doas,这里还会把权限设置为777(这里我增加了一个逻辑,如果设置了proxy,recursive也为true) /** boolean recursive = false; if (SessionState.get() != null) { recursive = (SessionState.get().isHiveServerQuery() && conf.getBoolean(HiveConf.ConfVars.HIVE_SERVER2_ENABLE_DOAS.varname, HiveConf.ConfVars.HIVE_SERVER2_ENABLE_DOAS.defaultBoolVal))||(HiveConf.getBoolVar(conf,HiveConf.ConfVars.HIVE_USE_CUSTOM_PROXY)); fsPermission = new FsPermission((short)00777); } */ origUmask = conf.get("fs.permissions.umask-mode"); conf.set("fs.permissions.umask-mode", "000"); } FileSystem fs = ShimLoader.getHadoopShims().getNonCachedFileSystem(mkdirPath.toUri(), conf); //这里是DFSClient的实例 boolean retval = false; try { retval = fs.mkdirs(mkdirPath, fsPermission); resetConfAndCloseFS(conf, recursive, origUmask, fs); } catch (IOException ioe) { try { resetConfAndCloseFS(conf, recursive, origUmask, fs); //调用resetConfAndCloseFS,reset fs.permissions.umask-mode的设置 } catch (IOException e) { // do nothing - double failure } } return retval; }
resetConfAndCloseFS方法用来重设fs.permissions.umask-mode的设置,这样如果后面创建目录不是使用Utilities.createDirsWithPermission就会使用这个重设的配置
private static void resetConfAndCloseFS (Configuration conf, boolean unsetUmask, String origUmask, FileSystem fs) throws IOException { if (unsetUmask) { //unsetUmask为true,即recursive为true的话,需要重设fs.permissions.umask-mode if (origUmask != null) { //如果有设置项的话,使用设置项 conf.set("fs.permissions.umask-mode", origUmask); } else { conf.unset("fs.permissions.umask-mode"); //这里虽然可以unset,后面会有默认值 } } fs.close(); }
通过查看DFSClient的源码,发现在DFSClient的构造函数中会初始化ugi的信息,默认为当前用户
final UserGroupInformation ugi; ... this.ugi = UserGroupInformation.getCurrentUser(); 如果更改成proxy用户,通过运行hadoop fs -mkdir测试,发现生成的文件目录属主还是当前登录用户 更改DFSClient的构造方法: //this.ugi = UserGroupInformation.getCurrentUser(); if(conf.getBoolean("use.custom.proxy",false)){ this.ugi = UserGroupInformation.createRemoteUser(conf.get("custom.proxy.user")); }else{ this.ugi = UserGroupInformation.getCurrentUser(); }
在hdfs-site.xml配置中增加:
dfs配置中增加:
use.custom.proxy true custom.proxy.user ericni
使用hdfs创建目录后,目录的属主仍然是hdfs,而数据写入的用户为提交job的用户。
因为上面的原因,要想使创建的hdfs的目录属主为proxy的用户,可以采用创建完后设置owner的方法。
通过查看DistributedFileSystem类的api,发现有setOwner的方法。
以insert overwrite 语句为例,在mapred job提交之前,会根据job的上下文内容,创建map和reduce的临时目录,这个目录是最终数据落地的目录,落地之后,在job完成的finally阶段,会通过MoveTask移动到对应的目录下面临时数据写入目录在ExecDriver类的execute方法中生成:
public int execute(DriverContext driverContext) { IOPrepareCache ioPrepareCache = IOPrepareCache.get(); ioPrepareCache.clear(); boolean success = true; Context ctx = driverContext.getCtx(); boolean ctxCreated = false; Path emptyScratchDir; MapWork mWork = work.getMapWork(); ReduceWork rWork = work.getReduceWork(); try { if (ctx == null) { ctx = new Context(job); ctxCreated = true; } emptyScratchDir = ctx.getMRTmpPath(); FileSystem fs = emptyScratchDir.getFileSystem(job); fs.mkdirs(emptyScratchDir); } catch (IOException e) { e.printStackTrace(); console.printError("Error launching map-reduce job", "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e)); return 5; } .... ListinputPaths = Utilities.getInputPaths(job, mWork, emptyScratchDir, ctx); //获取输入目录 Utilities.setInputPaths(job, inputPaths); Utilities.setMapRedWork(job, work, ctx.getMRTmpPath()); .... Utilities.createTmpDirs(job, mWork); //创建map临时目录 Utilities.createTmpDirs(job, rWork); //创建reduce临时目录
一种思路,在外层创建目录后setOwner,可以在Utilities中增加一个方法调用setOwner: public static void setDirWithOwner(Configuration conf,Path mkdirPath, String username,String groupname) throws IOException { LOG.warn("in Utilities setDirWithOwner path: " + mkdirPath + ",username: " + username + ",groupname: " + groupname); FileSystem fs = ShimLoader.getHadoopShims().getNonCachedFileSystem(mkdirPath.toUri(), conf); try { fs.setOwner(mkdirPath, username, groupname); //调用DistributedFileSystem.setOwner方法 }catch (IOException ios) { //no-op } }
同时更改createTmpDirs方法:
private static void createTmpDirs(Configuration conf, List> ops) throws IOException { FsPermission fsPermission = new FsPermission((short)00777); while (!ops.isEmpty()) { Operator extends OperatorDesc> op = ops.remove(0); if (op instanceof FileSinkOperator) { FileSinkDesc fdesc = ((FileSinkOperator) op).getConf(); //org.apache.hadoop.hive.ql.plan.FileSinkDesc Path tempDir = fdesc.getDirName(); //获取目录名 if (tempDir != null) { Path tempPath = Utilities.toTempPath(tempDir); //目录增加_tmp.前缀 createDirsWithPermission(conf, tempPath, fsPermission); if (conf.getBoolean("use.custom.proxy",false)) { //如果设置了use.custom.proxy,则调用setDirWithOwner方法,设置目录权限 LOG.warn("set owner after create dirs"); String username = conf.get("custom.proxy.user"); setDirWithOwner(conf,tempPath,username,null); } } } if (op.getChildOperators() != null) { ops.addAll(op.getChildOperators()); } } }
上面这种方法有一定的局限性,比如是使用了Utilities.createTmpDirs的方法创建的目录才有用(比如map或者reduce的临时数据目录)。
可以通过改下层的实现:
在DFSClient中增加一个setOwner方法:
public boolean setOwner(String src, String username) throws IOException { boolean setResult = false; checkOpen(); try { namenode.setOwner(src, username, null); setResult = true; } catch(RemoteException re) { throw re.unwrapRemoteException(AccessControlException.class, FileNotFoundException.class, SafeModeException.class, UnresolvedPathException.class); }finally{ return setResult; } }
同时更改primitiveMkdir为如下内容:
public boolean primitiveMkdir(String src, FsPermission absPermission, boolean createParent) throws IOException { checkOpen(); boolean MkRe; boolean SetRe; if (absPermission == null) { absPermission = FsPermission.getDefault().applyUMask(dfsClientConf.uMask); } if(LOG.isDebugEnabled()) { LOG.debug(src + ": masked=" + absPermission); } try { MkRe = namenode.mkdirs(src, absPermission, createParent); //namenode:org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolTranslatorPB if (this.conf.getBoolean("use.custom.proxy",false)){ LOG.warn("change primitiveMkdir add conf: " + this.conf.getBoolean("use.custom.proxy",false)); LOG.warn("change primitiveMkdir add conf: " + this.conf.get("custom.proxy.user")); String username = this.conf.get("custom.proxy.user"); if (("").equals(username)||username == null||("hdfs").equals(username)){ //no-op SetRe = true; }else{ SetRe = setOwner(src,username); } }else { SetRe = true; } return MkRe&&SetRe; } catch(RemoteException re) { throw re.unwrapRemoteException(AccessControlException.class, InvalidPathException.class, FileAlreadyExistsException.class, FileNotFoundException.class, ParentNotDirectoryException.class, SafeModeException.class, NSQuotaExceededException.class, DSQuotaExceededException.class, UnresolvedPathException.class); } }
这样,只要是调用了DFSClient的primitiveMkdir方法创建的目录(正常情况下创建目录都会调用primitiveMkdir方法),在proxy的情况下都可以更改目录。
到这里,hive的proxy算是开发完成了,为了实现proxy的功能,对hive和hadoop的代码更改如下:
1.HiveConf中增加两个配置项
2.重写HadoopDefaultAuthenticator的setConf方法
3.更改Context构造方法中关于scratch目录的项
4.更改Utilities中的createDirsWithPermission方法和createTmpDirs方法,并新增setDirWithOwner方法
5.更改HiveHistoryImpl构造方法中关于日志路径的项
6.更改JobClient的init方法
7.更改DFSClient的构造方法,增加一个setOwner方法,同时更改primitiveMkdir方法