Hadoop分布式文件系统的构架和设计(原创翻译 70%)
Replica Placement: The First Baby Steps 数据副本的存放: 婴儿的第一步
The placement of replicas is critical to HDFS reliability andperformance. Optimizing replica placement distinguishes HDFS from mostother distributed file systems. This is a feature that needs lots oftuning and experience. The purpose of a rack-aware replica placementpolicy is to improve data reliability, availability, and networkbandwidth utilization. The current implementation for the replicaplacement policy is a first effort in this direction. The short-termgoals of implementing this policy are to validate it on productionsystems, learn more about its behavior, and build a foundation to testand research more sophisticated policies.
The placement of replicas is critical to HDFS reliability andperformance. Optimizing replica placement distinguishes HDFS from mostother distributed file systems. This is a feature that needs lots oftuning and experience. The purpose of a rack-aware replica placementpolicy is to improve data reliability, availability, and networkbandwidth utilization. The current implementation for the replicaplacement policy is a first effort in this direction. The short-termgoals of implementing this policy are to validate it on productionsystems, learn more about its behavior, and build a foundation to testand research more sophisticated policies.
Large HDFS instances run on a cluster of computers that commonlyspread across many racks. Communication between two nodes in differentracks has to go through switches. In most cases, network bandwidthbetween machines in the same rack is greater than network bandwidthbetween machines in different racks.
Large HDFS instances run on a cluster of computers that commonlyspread across many racks. Communication between two nodes in differentracks has to go through switches. In most cases, network bandwidthbetween machines in the same rack is greater than network bandwidthbetween machines in different racks.
The NameNode determines the rack id each DataNode belongs to via the process outlined in Rack Awareness.A simple but non-optimal policy is to place replicas on unique racks.This prevents losing data when an entire rack fails and allows use ofbandwidth from multiple racks when reading data. This policy evenlydistributes replicas in the cluster which makes it easy to balance loadon component failure. However, this policy increases the cost of writesbecause a write needs to transfer blocks to multiple racks.
The NameNode determines the rack id each DataNode belongs to via the process outlined in Rack Awareness.A simple but non-optimal policy is to place replicas on unique racks.This prevents losing data when an entire rack fails and allows use ofbandwidth from multiple racks when reading data. This policy evenlydistributes replicas in the cluster which makes it easy to balance loadon component failure. However, this policy increases the cost of writesbecause a write needs to transfer blocks to multiple racks.
For the common case, when the replication factor is three, HDFS’splacement policy is to put one replica on one node in the local rack,another on a different node in the local rack, and the last on adifferent node in a different rack. This policy cuts the inter-rackwrite traffic which generally improves write performance. The chance ofrack failure is far less than that of node failure; this policy doesnot impact data reliability and availability guarantees. However, itdoes reduce the aggregate network bandwidth used when reading datasince a block is placed in only two unique racks rather than three.With this policy, the replicas of a file do not evenly distributeacross the racks. One third of replicas are on one node, two thirds ofreplicas are on one rack, and the other third are evenly distributedacross the remaining racks. This policy improves write performancewithout compromising data reliability or read performance.
For the common case, when the replication factor is three, HDFS’splacement policy is to put one replica on one node in the local rack,another on a different node in the local rack, and the last on adifferent node in a different rack. This policy cuts the inter-rackwrite traffic which generally improves write performance. The chance ofrack failure is far less than that of node failure; this policy doesnot impact data reliability and availability guarantees. However, itdoes reduce the aggregate network bandwidth used when reading datasince a block is placed in only two unique racks rather than three.With this policy, the replicas of a file do not evenly distributeacross the racks. One third of replicas are on one node, two thirds ofreplicas are on one rack, and the other third are evenly distributedacross the remaining racks. This policy improves write performancewithout compromising data reliability or read performance.
The current, default replica placement policy described here is a work in progress.
The current, default replica placement policy described here is a work in progress.
Replica Selection 复制选择
To minimize global bandwidth consumption and read latency, HDFStries to satisfy a read request from a replica that is closest to thereader. If there exists a replica on the same rack as the reader node,then that replica is preferred to satisfy the read request. If angg/HDFS cluster spans multiple data centers, then a replica that isresident in the local data center is preferred over any remote replica.
为了减少全局带宽和读延时, HDFS尝试把最近的一个副本给读的应用. 假如和读的应用在同一个机架存在副本, 则这个副本优先被读取. 假如HDFS群集存在多个数据中心, 则本地数据中心优先被读取.
SafeMode 安全模式
On startup, the Namenode enters a special state called Safemode.Replication of data blocks does not occur when the Namenode is in theSafemode state. The Namenode receives Heartbeat and Blockreportmessages from the Datanodes. A Blockreport contains the list of datablocks that a Datanode is hosting. Each block has a specified minimumnumber of replicas. A block is considered safely replicatedwhen the minimum number of replicas of that data block has checked inwith the Namenode. After a configurable percentage of safely replicateddata blocks checks in with the Namenode (plus an additional 30seconds), the Namenode exits the Safemode state. It then determines thelist of data blocks (if any) that still have fewer than the specifiednumber of replicas. The Namenode then replicates these blocks to otherDatanodes.
On startup, the Namenode enters a special state called Safemode.Replication of data blocks does not occur when the Namenode is in theSafemode state. The Namenode receives Heartbeat and Blockreportmessages from the Datanodes. A Blockreport contains the list of datablocks that a Datanode is hosting. Each block has a specified minimumnumber of replicas. A block is considered safely replicatedwhen the minimum number of replicas of that data block has checked inwith the Namenode. After a configurable percentage of safely replicateddata blocks checks in with the Namenode (plus an additional 30seconds), the Namenode exits the Safemode state. It then determines thelist of data blocks (if any) that still have fewer than the specifiednumber of replicas. The Namenode then replicates these blocks to otherDatanodes.
The Persistence of File System Metadata The Persistence of File System Metadata
The HDFS namespace is stored by the Namenode. The Namenode uses a transaction log called the EditLog to persistently record every change that occurs to file system metadata.For example, creating a new file in HDFS causes the Namenode to inserta record into the EditLog indicating this. Similarly, changing thereplication factor of a file causes a new record to be inserted intothe EditLog. The Namenode uses a file in its local host OSfile system to store the EditLog. The entire file system namespace,including the mapping of blocks to files and file system properties, isstored in a file called the FsImage. The FsImage is stored as a file in the Namenode’s local file system too.
The HDFS namespace is stored by the Namenode. The Namenode uses a transaction log called the EditLog to persistently record every change that occurs to file system metadata.For example, creating a new file in HDFS causes the Namenode to inserta record into the EditLog indicating this. Similarly, changing thereplication factor of a file causes a new record to be inserted intothe EditLog. The Namenode uses a file in its local host OSfile system to store the EditLog. The entire file system namespace,including the mapping of blocks to files and file system properties, isstored in a file called the FsImage. The FsImage is stored as a file in the Namenode’s local file system too.
The Namenode keeps an image of the entire file system namespace and file Blockmapin memory. This key metadata item is designed to be compact, such thata Namenode with 4 GB of RAM is plenty to support a huge number of filesand directories. When the Namenode starts up, it reads the FsImage andEditLog from disk, applies all the transactions from the EditLog to thein-memory representation of the FsImage, and flushes out this newversion into a new FsImage on disk. It can then truncate the oldEditLog because its transactions have been applied to the persistentFsImage. This process is called a checkpoint. In the currentimplementation, a checkpoint only occurs when the Namenode starts up.Work is in progress to support periodic checkpointing in the nearfuture.
The Namenode keeps an image of the entire file system namespace and file Blockmapin memory. This key metadata item is designed to be compact, such thata Namenode with 4 GB of RAM is plenty to support a huge number of filesand directories. When the Namenode starts up, it reads the FsImage andEditLog from disk, applies all the transactions from the EditLog to thein-memory representation of the FsImage, and flushes out this newversion into a new FsImage on disk. It can then truncate the oldEditLog because its transactions have been applied to the persistentFsImage. This process is called a checkpoint. In the currentimplementation, a checkpoint only occurs when the Namenode starts up.Work is in progress to support periodic checkpointing in the nearfuture.
The Datanode stores HDFS data in files in its local file system. TheDatanode has no knowledge about HDFS files. It stores each block ofHDFS data in a separate file in its local file system. The Datanodedoes not create all files in the same directory. Instead, it uses aheuristic to determine the optimal number of files per directory andcreates subdirectories appropriately. It is not optimal to create alllocal files in the same directory because the local file system mightnot be able to efficiently support a huge number of files in a singledirectory. When a Datanode starts up, it scans through its local filesystem, generates a list of all HDFS data blocks that correspond toeach of these local files and sends this report to the Namenode: thisis the Blockreport.
The Datanode stores HDFS data in files in its local file system. TheDatanode has no knowledge about HDFS files. It stores each block ofHDFS data in a separate file in its local file system. The Datanodedoes not create all files in the same directory. Instead, it uses aheuristic to determine the optimal number of files per directory andcreates subdirectories appropriately. It is not optimal to create alllocal files in the same directory because the local file system mightnot be able to efficiently support a huge number of files in a singledirectory. When a Datanode starts up, it scans through its local filesystem, generates a list of all HDFS data blocks that correspond toeach of these local files and sends this report to the Namenode: thisis the Blockreport.
The Communication Protocols 通讯协议
All HDFS communication protocols are layered on top of the TCP/IPprotocol. A client establishes a connection to a configurable TCP port on the Namenode machine. It talks the ClientProtocol with the Namenode. The Datanodes talk to the Namenode using the DatanodeProtocol. A Remote Procedure Call (RPC)abstraction wraps both the ClientProtocol and the DatanodeProtocol. Bydesign, the Namenode never initiates any RPCs. Instead, it onlyresponds to RPC requests issued by Datanodes or clients.
All HDFS communication protocols are layered on top of the TCP/IPprotocol. A client establishes a connection to a configurable TCP port on the Namenode machine. It talks the ClientProtocol with the Namenode. The Datanodes talk to the Namenode using the DatanodeProtocol. A Remote Procedure Call (RPC)abstraction wraps both the ClientProtocol and the DatanodeProtocol. Bydesign, the Namenode never initiates any RPCs. Instead, it onlyresponds to RPC requests issued by Datanodes or clients.
Robustness 健壮性
The primary objective of HDFS is to store data reliably even in thepresence of failures. The three common types of failures are Namenodefailures, Datanode failures and network partitions.
The primary objective of HDFS is to store data reliably even in thepresence of failures. The three common types of failures are Namenodefailures, Datanode failures and network partitions.
Data Disk Failure, Heartbeats and Re-Replication 磁盘错误, 心跳 and 再复制
Each Datanode sends a Heartbeat message to the Namenodeperiodically. A network partition can cause a subset of Datanodes tolose connectivity with the Namenode. The Namenode detects thiscondition by the absence of a Heartbeat message. The Namenode marksDatanodes without recent Heartbeats as dead and does not forward anynew IOrequests to them. Any data that was registered to a dead Datanode isnot available to HDFS any more. Datanode death may cause thereplication factor of some blocks to fall below their specified value.The Namenode constantly tracks which blocks need to be replicated andinitiates replication whenever necessary. The necessity forre-replication may arise due to many reasons: a Datanode may becomeunavailable, a replica may become corrupted, a hard disk on a Datanodemay fail, or the replication factor of a file may be increased.
Each Datanode sends a Heartbeat message to the Namenodeperiodically. A network partition can cause a subset of Datanodes tolose connectivity with the Namenode. The Namenode detects thiscondition by the absence of a Heartbeat message. The Namenode marksDatanodes without recent Heartbeats as dead and does not forward anynew IOrequests to them. Any data that was registered to a dead Datanode isnot available to HDFS any more. Datanode death may cause thereplication factor of some blocks to fall below their specified value.The Namenode constantly tracks which blocks need to be replicated andinitiates replication whenever necessary. The necessity forre-replication may arise due to many reasons: a Datanode may becomeunavailable, a replica may become corrupted, a hard disk on a Datanodemay fail, or the replication factor of a file may be increased.
Cluster Rebalancing 全局负载平衡
The HDFS architecture is compatible with data rebalancing schemes.A scheme might automatically move data from one Datanode to another ifthe free space on a Datanode falls below a certain threshold. In theevent of a sudden high demand for a particular file, a scheme mightdynamically create additional replicas and rebalance other data in thecluster. These types of data rebalancing schemes are not yetimplemented.
The HDFS architecture is compatible with data rebalancing schemes.A scheme might automatically move data from one Datanode to another ifthe free space on a Datanode falls below a certain threshold. In theevent of a sudden high demand for a particular file, a scheme mightdynamically create additional replicas and rebalance other data in thecluster. These types of data rebalancing schemes are not yetimplemented.
Data Integrity 数据完整性
<!-- --> It ispossible that a block of data fetched from a Datanode arrivescorrupted. This corruption can occur because of faults in a storagedevice, network faults, or buggy software. The HDFS client softwareimplements checksum checking on the contents of HDFS files. When aclient creates an HDFS file, it computes a checksum of each block ofthe file and stores these checksums in a separate hidden file in thesame HDFS namespace. When a client retrieves file contents it verifiesthat the data it received from each Datanode matches the checksumstored in the associated checksum file. If not, then the client can optto retrieve that block from another Datanode that has a replica of thatblock.
<!-- --> It ispossible that a block of data fetched from a Datanode arrivescorrupted. This corruption can occur because of faults in a storagedevice, network faults, or buggy software. The HDFS client softwareimplements checksum checking on the contents of HDFS files. When aclient creates an HDFS file, it computes a checksum of each block ofthe file and stores these checksums in a separate hidden file in thesame HDFS namespace. When a client retrieves file contents it verifiesthat the data it received from each Datanode matches the checksumstored in the associated checksum file. If not, then the client can optto retrieve that block from another Datanode that has a replica of thatblock.
Metadata Disk Failure 元数据磁盘故障
The FsImage and the EditLog are central data structures of HDFS. Acorruption of these files can cause the HDFS instance to benon-functional. For this reason, the Namenode can be configured tosupport maintaining multiple copies of the FsImage and EditLog. Anyupdate to either the FsImage or EditLog causes each of the FsImages andEditLogs to get updated synchronously. This synchronous updating ofmultiple copies of the FsImage and EditLog may degrade the rate ofnamespace transactions per second that a Namenode can support. However,this degradation is acceptable because even though HDFS applicationsare very data intensive in nature, they are not metadata intensive. When a Namenode restarts, it selects the latest consistent FsImage and EditLog to use.
FsImage和EditLog是HDFS中心的数据结构. 这些文件中一个损毁阿会引起HDFS实例的不能正常工作. 因为这个原因,名字节点能够被设置为支持维护多个FsImage和EditLog的副本.任何一个FsImage或EditLog更新了，引起其他的FsImages和EditLogs都同步更新了.这个同步更新FsImage、EditLog多个copy的机制，会减少名字节点每秒处理事务的数量. 无论如何,这个损失是可以被接受的，因为即使HDFS应用程序的运算速度是非常重要的，但也没有元数据重要. 当名字节点重起,它选择最新，且数据一致的FsImage和EditLog被使用.
The Namenode machine is a single point of failure for an HDFScluster. If the Namenode machine fails, manual intervention isnecessary. Currently, automatic restart and failover of the Namenodesoftware to another machine is not supported.
名字节点服务器是HDFS群集中的一个单点故障点. 假如名字节点失效了, 人工的操作是必须的. 在当前, 自动重起并且修复错误，自动将名字节点软件部署到另外一台机器还没有被支持。
Snapshots 数据快照
Snapshots support storing a copy of data at a particular instant oftime. One usage of the snapshot feature may be to roll back a corruptedHDFS instance to a previously known good point in time. HDFS does notcurrently support snapshots but will in a future release.
快照支持存储一个分布式文件系统某个时间的一份copy数据。快照的用处是可以回滚HDFS实例到之前好的状态点。HDFS现在还不支持快照，但是以后版本打算支持.
Data Organization 数据组织 Data Blocks 数据块
Data Blocks 数据块
HDFS is designed to support very large files. Applications that arecompatible with HDFS are those that deal with large data sets. Theseapplications write their data only once but they read it one or moretimes and require these reads to be satisfied at streaming speeds. HDFSsupports write-once-read-many semantics on files. A typical block sizeused by HDFS is 64 MB. Thus, an HDFS file is chopped up into 64 MBchunks, and if possible, each chunk will reside on a differentDatanode.
HDFS被设计为支持非常大的文件. 在HDFS运行的软件都是处理大数据集的.这些应用程序一般写一次数据，但是可能需要顺畅的对那些数据读一次或多次. HDFS支持写一次读多次的文件语义.一个典型的HDFS文件块大小是64MB. 应次, 一个HDFS文件被分割成64MB大小的数据块集合, 如果可能,每一个块可以在不同的数据节点上。
Staging 分段运输
A client request to create a file does not reach the Namenodeimmediately. In fact, initially the HDFS client caches the file datainto a temporary local file. Application writes are transparentlyredirected to this temporary local file. When the local fileaccumulates data worth over one HDFS block size, the client contactsthe Namenode. The Namenode inserts the file name into the file systemhierarchy and allocates a data block for it. The Namenode responds tothe client request with the identity of the Datanode and thedestination data block. Then the client flushes the block of data fromthe local temporary file to the specified Datanode. When a file isclosed, the remaining un-flushed data in the temporary local file istransferred to the Datanode. The client then tells the Namenode thatthe file is closed. At this point, the Namenode commits the filecreation operation into a persistent store. If the Namenode dies beforethe file is closed, the file is lost.
A client request to create a file does not reach the Namenodeimmediately. In fact, initially the HDFS client caches the file datainto a temporary local file. Application writes are transparentlyredirected to this temporary local file. When the local fileaccumulates data worth over one HDFS block size, the client contactsthe Namenode. The Namenode inserts the file name into the file systemhierarchy and allocates a data block for it. The Namenode responds tothe client request with the identity of the Datanode and thedestination data block. Then the client flushes the block of data fromthe local temporary file to the specified Datanode. When a file isclosed, the remaining un-flushed data in the temporary local file istransferred to the Datanode. The client then tells the Namenode thatthe file is closed. At this point, the Namenode commits the filecreation operation into a persistent store. If the Namenode dies beforethe file is closed, the file is lost.
The above approach has been adopted after careful consideration oftarget applications that run on HDFS. These applications need streamingwrites to files. If a client writes to a remote file directly withoutany client side buffering, the network speed and the congestion in thenetwork impacts throughput considerably. This approach is not withoutprecedent. Earlier distributed file systems, e.g. AFS,have used client side caching to improve performance. A POSIXrequirement has been relaxed to achieve higher performance of datauploads.
The above approach has been adopted after careful consideration oftarget applications that run on HDFS. These applications need streamingwrites to files. If a client writes to a remote file directly withoutany client side buffering, the network speed and the congestion in thenetwork impacts throughput considerably. This approach is not withoutprecedent. Earlier distributed file systems, e.g. AFS,have used client side caching to improve performance. A POSIXrequirement has been relaxed to achieve higher performance of datauploads.
Replication Pipelining 管道方式的复制操作
When a client is writing data to an HDFS file, its data is firstwritten to a local file as explained in the previous section. Supposethe HDFS file has a replication factor of three. When the local fileaccumulates a full block of user data, the client retrieves a list ofDatanodes from the Namenode. This list contains the Datanodes that willhost a replica of that block. The client then flushes the data block tothe first Datanode. The first Datanode starts receiving the data insmall portions (4 KB), writes each portion to its local repository andtransfers that portion to the second Datanode in the list. The secondDatanode, in turn starts receiving each portion of the data block,writes that portion to its repository and then flushes that portion tothe third Datanode. Finally, the third Datanode writes the data to itslocal repository. Thus, a Datanode can be receiving data from theprevious one in the pipeline and at the same time forwarding data tothe next one in the pipeline. Thus, the data is pipelined from oneDatanode to the next.
当一个客户端写数据到HDFS文件时,数据前面一段先写入本地文件. 假设，HDFS文件的副本参数为3.当本地文件累计到满一个数据块时,客户端从名字节点得到一个数据节点列表.这些数据节点将存放这个数据块的一个副本.接着，客户端刷新数据到第一个数据节点. 第一个数据节点开始接收数据,一小块一小块接收（4K）,将每一小块的数据写到本地存储，同时将这一小块数据传输到列表上的第二个数据节点上. 第二个数据节点,继续接受数据写到本地存储，接着传输到第三个数据节点上。最后第三节点将数据写到它的本地存储上。就这样，一个数据节点能够从管道的前一个接收数据，同时又将数据传给管道中的下一个节点，就这样数据在管道中从一个数据节点传送到另一个数据节点。
Accessibility 访问方式
HDFS can be accessed from applications in many different ways. Natively, HDFS provides a Java APIfor applications to use. A C language wrapper for this Java API is alsoavailable. In addition, an HTTP browser can also be used to browse thefiles of an HDFS instance. Work is in progress to expose HDFS throughthe WebDAV protocol.
应用能够通过多总方式访问HDFS. 原生接口, HDFS提供Java 应用程序接口. C语言包装的Java 应用程序接口. 另外, 浏览器能够HDFS上的文件. Work is in progress to expose HDFS through the WebDAV protocol.
DFSShell 分布式文件系统命令行接口
HDFS allows user data to be organized in the form of files and directories. It provides a commandline interface called DFSShellthat lets a user interact with the data in HDFS. The syntax of thiscommand set is similar to other shells (e.g. bash, csh) that users arealready familiar with. Here are some sample action/command pairs:
HDFS允许用户数据被组织成文件和目录的形式. 提供的命令行形式的接口叫DFSShell，是用户和HDFS中数据交互的一种接口. 语法有点像其他用户已经熟悉的命令行环境(例如 bash, csh). 这里提供一些功能和命令的例子:
Action 功能 Command 命令 Create a directory named /foodir bin/hadoop dfs -mkdir /foodir 建立一个目录 /foodir bin/hadoop dfs -mkdir /foodir View the contents of a file named /foodir/myfile.txt bin/hadoop dfs -cat /foodir/myfile.txt 查看/foodir/myfile.txt文件的内容 bin/hadoop dfs -cat /foodir/myfile.txtDFSShell is targeted for applications that need a scripting language to interact with the stored data.
DFSShell的目的是为了应用程序通过脚本访问HDFS中的数据.
DFSAdmin 管理工具
The DFSAdmin command set is used foradministering an HDFS cluster. These are commands that are used only byan HDFS administrator. Here are some sample action/command pairs:
DFSAdmin的一组命令是用于管理HDFS群集. 这些命令主要给HDFS管理员使用. 这里提供一些功能和命令的例子:
Action 功能 Command 命令 Put a cluster in SafeMode （设置群集进入安全模式） bin/hadoop dfsadmin -safemode enter Generate a list of Datanodes (产生一个数据节点列表) bin/hadoop dfsadmin -report Decommission Datanode datanodename bin/hadoop dfsadmin -decommission datanodename 使数据节点datanodename推出 bin/hadoop dfsadmin -decommission datanodename Browser Interface 浏览器接口
A typical HDFS install configures a web server to expose the HDFSnamespace through a configurable TCP port. This allows a user tonavigate the HDFS namespace and view the contents of its files using aweb browser.
一个典型的HDFS安装通过一个可配置的端口的网站服务器来暴露HDFS名字空间. 他允许用户浏览HDFS名字空间和浏览通过浏览器浏览文件.
Space Reclamation 空间的回收 File Deletes and Undeletes 文件的删除和恢复
File Deletes and Undeletes 文件的删除和恢复
When a file is deleted by a user or an application, it is notimmediately removed from HDFS. Instead, HDFS first renames it to a filein the /trash directory. The file can be restored quickly as long as it remains in /trash. A file remains in /trash for a configurable amount of time. After the expiry of its life in /trash,the Namenode deletes the file from the HDFS namespace. The deletion ofa file causes the blocks associated with the file to be freed. Notethat there could be an appreciable time delay between the time a fileis deleted by a user and the time of the corresponding increase in freespace in HDFS.
当一个文件被用户删除，它没有立即被HDFS文件系统删除. HDFS先把它改名到/trash目录.文件只要在 /trash中，就能被快速的恢复. 文件在 /trash保留一定的时间，是可以配置的. 当超过了/trash的生命周期, 名字服务器将会删除这个文件. 然后文件的空间被释放. 文件的删除到HDFS存储空间的增加会有一些延时.
A user can Undelete a file after deleting it as long as it remains in the /trash directory. If a user wants to undelete a file that he/she has deleted, he/she can navigate the /trash directory and retrieve the file. The /trash directory contains only the latest copy of the file that was deleted. The /trashdirectory is just like any other directory with one special feature:HDFS applies specified policies to automatically delete files from thisdirectory. The current default policy is to delete files from /trash that are more than 6 hours old. In the future, this policy will be configurable through a well defined interface.
只要文件在/trash目录中，文件就能被恢复. 用户如果想恢复/trash目录中的文件，只需直接访问/trash这个路径。/trash目录仅仅包含最近删除文件的copy. /trash和其他文件一样仅仅多了一个特性: HDFS有一个自动删除其中文件的策略. 当前的策略是，删除的文件在/trash中，保留6个小时. 以后，这个策略将会通过一个良好的接口（配置文件）配置.
Decrease Replication Factor 减少复制因子
When the replication factor of a file is reduced, the Namenode selectsexcess replicas that can be deleted. The next Heartbeat transfers thisinformation to the Datanode. The Datanode then removes thecorresponding blocks and the corresponding free space appears in t