org.apache.hadoop.filecache
Class TrackerDistributedCacheManager

java.lang.Object
  org.apache.hadoop.filecache.TrackerDistributedCacheManager

public class TrackerDistributedCacheManager
extends Object

Manages a single machine's instance of a cross-job cache. This class would typically be instantiated by a TaskTracker (or something that emulates it, like LocalJobRunner). This class is internal to Hadoop, and should not be treated as a public interface.

Nested Class Summary

protected class	TrackerDistributedCacheManager.BaseDirManager This class holds properties of each base directories and is responsible for clean up unused cache files in base directories.
protected class	TrackerDistributedCacheManager.CleanupThread A thread to check and cleanup the unused files periodically

Field Summary

protected TrackerDistributedCacheManager.BaseDirManager	baseDirManager
protected TrackerDistributedCacheManager.CleanupThread	cleanupThread

Constructor Summary

TrackerDistributedCacheManager(Configuration conf, TaskController controller)

Method Summary

static void	createAllSymlink(Configuration conf, File jobCacheDir, File workDir) This method create symlinks for all files in a given dir in another directory.
static void	determineCacheVisibilities(Configuration job) Determines the visibilities of the distributed cache files and archives.
static void	determineTimestamps(Configuration job) Determines timestamps of files to be cached, and stores those in the configuration.
static long	downloadCacheObject(Configuration conf, URI source, Path destination, long desiredTimestamp, boolean isArchive, FsPermission permission) Download a given path to the local file system.
static boolean[]	getArchiveVisibilities(Configuration conf) Get the booleans on whether the archives are public or not.
static void	getDelegationTokens(Configuration job, Credentials credentials) For each archive or cache file - get the corresponding delegation token
static boolean[]	getFileVisibilities(Configuration conf) Get the booleans on whether the files are public or not.
protected TaskDistributedCacheManager	getTaskDistributedCacheManager(JobID jobId)
TaskDistributedCacheManager	newTaskDistributedCacheManager(JobID jobId, Configuration taskConf)
void	purgeCache() Clear the entire contents of the cache and delete the backing files.
void	removeTaskDistributedCacheManager(JobID jobId)
void	setArchiveSizes(JobID jobId, long[] sizes) Set the sizes for any archives, files, or directories in the private distributed cache.
void	startCleanupThread() Start the background thread
void	stopCleanupThread() Stop the background thread
static void	validate(Configuration conf) This is part of the framework API.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

baseDirManager

protected TrackerDistributedCacheManager.BaseDirManager baseDirManager

cleanupThread

protected TrackerDistributedCacheManager.CleanupThread cleanupThread

Constructor Detail

TrackerDistributedCacheManager

public TrackerDistributedCacheManager(Configuration conf,
                                      TaskController controller)
                               throws IOException

Throws:

IOException

Method Detail

downloadCacheObject

public static long downloadCacheObject(Configuration conf,
                                       URI source,
                                       Path destination,
                                       long desiredTimestamp,
                                       boolean isArchive,
                                       FsPermission permission)
                                throws IOException

Download a given path to the local file system.

Parameters:

conf - the job's configuration

source - the source to copy from

destination - where to copy the file. must be local fs

desiredTimestamp - the required modification timestamp of the source

isArchive - is this an archive that should be expanded

permission - the desired permissions of the file.

Returns:

for archives, the number of bytes in the unpacked directory

Throws:

IOException

createAllSymlink

public static void createAllSymlink(Configuration conf,
                                    File jobCacheDir,
                                    File workDir)
                             throws IOException

This method create symlinks for all files in a given dir in another directory. Should not be used outside of DistributedCache code.

Parameters:

conf - the configuration

jobCacheDir - the target directory for creating symlinks

workDir - the directory in which the symlinks are created

Throws:

IOException

purgeCache

public void purgeCache()

Clear the entire contents of the cache and delete the backing files. This should only be used when the server is reinitializing, because the users are going to lose their files.

newTaskDistributedCacheManager

public TaskDistributedCacheManager newTaskDistributedCacheManager(JobID jobId,
                                                                  Configuration taskConf)
                                                           throws IOException

Throws:

IOException

setArchiveSizes

public void setArchiveSizes(JobID jobId,
                            long[] sizes)
                     throws IOException

Set the sizes for any archives, files, or directories in the private distributed cache.

Throws:

IOException

removeTaskDistributedCacheManager

public void removeTaskDistributedCacheManager(JobID jobId)

getTaskDistributedCacheManager

protected TaskDistributedCacheManager getTaskDistributedCacheManager(JobID jobId)

determineTimestamps

public static void determineTimestamps(Configuration job)
                                throws IOException

Determines timestamps of files to be cached, and stores those in the configuration. This is intended to be used internally by JobClient after all cache files have been added. This is an internal method!

Parameters:

job - Configuration of a job.

Throws:

IOException

determineCacheVisibilities

public static void determineCacheVisibilities(Configuration job)
                                       throws IOException

Determines the visibilities of the distributed cache files and archives. The visibility of a cache path is "public" if the leaf component has READ permissions for others, and the parent subdirs have EXECUTE permissions for others

Parameters:

job -

Throws:

IOException

getFileVisibilities

public static boolean[] getFileVisibilities(Configuration conf)

Get the booleans on whether the files are public or not. Used by internal DistributedCache and MapReduce code.

Parameters:

conf - The configuration which stored the timestamps

Returns:

array of booleans

Throws:

IOException

getArchiveVisibilities

public static boolean[] getArchiveVisibilities(Configuration conf)

Get the booleans on whether the archives are public or not. Used by internal DistributedCache and MapReduce code.

Parameters:

conf - The configuration which stored the timestamps

Returns:

array of booleans

getDelegationTokens

public static void getDelegationTokens(Configuration job,
                                       Credentials credentials)
                                throws IOException

For each archive or cache file - get the corresponding delegation token

Parameters:

job -

credentials -

Throws:

IOException

validate

public static void validate(Configuration conf)
                     throws InvalidJobConfException

This is part of the framework API. It's called within the job submission code only, not by users. In the non-error case it has no side effects and returns normally. If there's a URI in both mapred.cache.files and mapred.cache.archives, it throws its exception.

Parameters:

conf - a Configuration to be cheked for duplication in cached URIs

Throws:

InvalidJobConfException

startCleanupThread

public void startCleanupThread()

Start the background thread

stopCleanupThread

public void stopCleanupThread()

Stop the background thread