hive在hdfs中的job中间文件是根据当前登陆用户产生的，其默认值为/tmp/hive-${user.name}，这就导致实现proxy的功能时会遇到临时文件的权限问题，比如在实现了proxy功能后，以超级用户hdfs proxy到普通用户user时，在hdfs中的临时文件在/tmp/hive-user目录中，而目录的属主是hdfs,这时再以普通用户user运行job时，对这个目录就会有权限问题，下面说下这里proxy的实现和解决权限问题的方法：
1.实现proxy功能
更改org.apache.hadoop.hive.ql.Context类的构造方法：
public Context(Configuration conf, String executionId) {
this.conf = conf;
this.executionId = executionId;
if(HiveConf.getBoolVar(conf,HiveConf.ConfVars.HIVE_USE_CUSTOM_PROXY)){
String proxyUser = HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_CUSTOM_PROXY_USER);
LOG.warn("use custom proxy,gen Scratch path,proxy user is " + proxyUser);
if(("").equals(proxyUser)||proxyUser == null||("hdfs").equals(proxyUser)){
nonLocalScratchPath = new Path(HiveConf.getVar(conf, HiveConf.ConfVars.SCRATCHDIR),executionId);
localScratchDir = new Path(HiveConf.getVar(conf, HiveConf.ConfVars.LOCALSCRATCHDIR),executionId).toUri().getPath();
}else{
localScratchDir = new Path((System.getProperty("java.io.tmpdir") + File.separator + proxyUser),executionId).toUri().getPath();
nonLocalScratchPath = new Path(("/tmp/hive-" + proxyUser),executionId);
}
}else{
nonLocalScratchPath = new Path(HiveConf.getVar(conf, HiveConf.ConfVars.SCRATCHDIR),executionId);
localScratchDir = new Path(HiveConf.getVar(conf, HiveConf.ConfVars.LOCALSCRATCHDIR),executionId).toUri().getPath();
}
LOG.warn("in Context init function nonLocalScratchPath is " + nonLocalScratchPath);
LOG.warn("in Context init function localScratchPath is " + localScratchDir);
scratchDirPermission= HiveConf.getVar(conf, HiveConf.ConfVars.SCRATCHDIRPERMISSION);
}2.权限问题的解决
在上面的代码中可以看到scratchDirPermission的设置，这个是指创建的目录的权限，默认是700，因为是中间文件，我们可以把权限设置的大一点，比如777，在设置了777之后，却发现目录的权限是755.
根据报错的堆栈可以看到方法在Context中调用的情况:
getExternalTmpPath--->getExternalScratchDir-->getScratchDir-->Utilities.createDirsWithPermission
(目录不存在时，根据HiveConf.ConfVars.SCRATCHDIRPERMISSION的设置创建hdfs tmp目录)
看下getScratchDir方法：
private final Map<String, Path> fsScratchDirs = new HashMap<String, Path>();
.....
private Path getScratchDir(String scheme, String authority,
boolean mkdir, String scratchDir) { // 如果是explain语句mkdir为false
String fileSystem = scheme + ":" + authority;
Path dir = fsScratchDirs.get(fileSystem + "-" + TaskRunner.getTaskRunnerID());
if (dir == null) {
Path dirPath = new Path(scheme, authority,
scratchDir + "-" + TaskRunner.getTaskRunnerID());
if (mkdir) {
try {
FileSystem fs = dirPath.getFileSystem( conf);
dirPath = new Path(fs.makeQualified(dirPath).toString());
FsPermission fsPermission = new FsPermission(Short.parseShort(scratchDirPermission.trim(), 8));
// 目录权限由HiveConf.ConfVars.SCRATCHDIRPERMISSION 设置
if (!Utilities.createDirsWithPermission(conf , dirPath, fsPermission)) {
throw new RuntimeException("Cannot make directory: "
+ dirPath.toString());
}
if (isHDFSCleanup ) {
fs.deleteOnExit(dirPath);
}
} catch (IOException e) {
throw new RuntimeException (e);
}
}
dir = dirPath;
fsScratchDirs.put(fileSystem + "-" + TaskRunner.getTaskRunnerID(), dir);
}
return dir;
}调用Utilities.createDirsWithPermission方法时，传入的目录的权限（由HiveConf.ConfVars.SCRATCHDIRPERMISSION 设置）默认是700
org.apache.hadoop.hive.ql.exec.Utilities类的createDirsWithPermission方法内容如下：
public static boolean createDirsWithPermission(Configuration conf, Path mkdir,
FsPermission fsPermission) throws IOException {
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);
//如果是来自hiverserver的请求，并且开启了doas，recursive为true,权限设置为777,umask 000
fsPermission = new FsPermission((short)00777);
}
// if we made it so far without exception we are good!
return createDirsWithPermission(conf, mkdir, fsPermission, recursive); //默认recursive为false
}
.....
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,origUmask 为000，否则为null
origUmask = conf.get("fs.permissions.umask-mode");
// this umask is required because by default the hdfs mask is 022 resulting in
// all parents getting the fsPermission & !(022) permission instead of fsPermission
conf.set("fs.permissions.umask-mode", "000");
}
FileSystem fs = ShimLoader.getHadoopShims().getNonCachedFileSystem(mkdirPath.toUri(), conf);
LOG.warn("fs.permissions.umask-mode is " + conf.get("fs.permissions.umask-mode")); //默认为022
boolean retval = false;
try {
retval = fs.mkdirs(mkdirPath, fsPermission);
resetConfAndCloseFS(conf, recursive, origUmask, fs);
//这里因为recursive为false,导致不会重置fs.permissions.umask-mode的配置，
即fs.permissions.umask-mode为022，因此导致即使设置了权限为777，创建的目录权限最终还是为755
} catch (IOException ioe) {
try {
resetConfAndCloseFS(conf, recursive, origUmask, fs);
}
catch (IOException e) {
// do nothing - double failure
}
}
return retval;
}hdfs中关于fs.permissions.umask-mode的配置，默认是002
public static final String FS_PERMISSIONS_UMASK_KEY = "fs.permissions.umask-mode"; public static final int FS_PERMISSIONS_UMASK_DEFAULT = 0022;
为了实现可以创建777权限的临时文件目录，更改createDirsWithPermission方法如下：
public static boolean createDirsWithPermission(Configuration conf, Path mkdir,
FsPermission fsPermission) throws IOException {
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);
}
// if we made it so far without exception we are good!
return createDirsWithPermission(conf, mkdir, fsPermission, recursive);
}这样，就可以创建出777的hdfs目录了。
本文出自 “菜光光的博客” 博客，请务必保留此出处http://caiguangguang.blog.51cto.com/1652935/1589879
原文地址：http://caiguangguang.blog.51cto.com/1652935/1589879
