首页 > 其他好文 > 详细

【转】Hadoop安全模式详解及配置

时间：2015-10-03 14:20:46 阅读：232 评论：0 收藏：0 [点我收藏+]

标签：

原文链接 http://www.iteblog.com/archives/977

　在《Hadoop 1.x中fsimage和edits合并实现》文章中提到，Hadoop的NameNode在重启的时候，将会进入到安全模式。而在安全模式，HDFS只支持访问元数据的操作才会返回成功，其他的操作诸如创建、删除文件等操作都会导致失败。　　NameNode在重启的时候，DataNode需要向NameNode发送块的信息，NameNode只有获取到整个文件系统中有99.9%（可以配置的）的块满足最小副本才会自动退出安全模式。最小副本和那个99.9%阀值可以通过下面配置来设定：

01
<property>

02
  <name>dfs.namenode.replication.min</name>

03
  <value>1</value>

04
  <description>Minimal block replication.

05
  </description>

06
</property>

07
 
08
<property>

09
  <name>dfs.namenode.safemode.threshold-pct</name>

10
  <value>0.999f</value>

11
  <description>

12
    Specifies the percentage of blocks that should satisfy the minimal

13
    replication requirement defined by dfs.namenode.replication.min.

14
    Values less than or equal to 0 mean not to wait for any particular

15
    percentage of blocks before exiting safemode.

16
    Values greater than 1 will make safe mode permanent.

17
  </description>

18
</property>

　　Hadoop中每个块的默认最小副本为1；dfs.namenode.safemode.threshold-pct参数的意思是指定达到最小副本数的数据块的百分比。这个值小等于0表示无须等待就可以退出安全模式；而如果这个值大于1表示永远处于安全模式。除了上面两个参数对安全模式有影响之外，下面几个参数也会对安全模式有影响

01
<property>

02
  <name>dfs.namenode.safemode.min.datanodes</name>

03
  <value>0</value>

04
  <description>

05
    Specifies the number of datanodes that must be considered alive

06
    before the name node exits safemode.

07
    Values less than or equal to 0 mean not to take the number of live

08
    datanodes into account when deciding whether to remain in safe mode

09
    during startup.

10
    Values greater than the number of datanodes in the cluster

11
    will make safe mode permanent.

12
  </description>

13
</property>

14
 
15
<property>

16
  <name>dfs.namenode.safemode.extension</name>

17
  <value>30000</value>

18
  <description>

19
    Determines extension of safe mode in milliseconds

20
    after the threshold level is reached.

21
  </description>

22
</property>

　　dfs.namenode.safemode.min.datanodes的意思指namenode退出安全模式之前有效的（活着的）datanode的数量。这个值小等于0表示在退出安全模式之前无须考虑有效的datanode节点个数，值大于集群中datanode节点总数则表示永远处于安全模式；dfs.namenode.safemode.extension表示在满足dfs.namenode.safemode.threshold-pct值之后，NameNode还需要处于安全模式的时间（单位是秒）。来看看代码是怎么弄的：

01
/**

02
 * Creates SafeModeInfo when the name node enters

03
 * automatic safe mode at startup.

04
 * 

05
 * @param conf configuration

06
 */

07
private SafeModeInfo(Configuration conf) {

08
      this.threshold = conf.getFloat(DFS_NAMENODE_SAFEMODE_THRESHOLD_PCT_KEY,

09
          DFS_NAMENODE_SAFEMODE_THRESHOLD_PCT_DEFAULT);

10
      if(threshold > 1.0) {

11
        LOG.warn("The threshold value should‘t be greater

12
                                          than 1, threshold: " + threshold);

13
      }

14
      this.datanodeThreshold = conf.getInt(

15
        DFS_NAMENODE_SAFEMODE_MIN_DATANODES_KEY,

16
        DFS_NAMENODE_SAFEMODE_MIN_DATANODES_DEFAULT);

17
      this.extension = conf.getInt(DFS_NAMENODE_SAFEMODE_EXTENSION_KEY, 0);

18
      this.safeReplication =conf.getInt(DFS_NAMENODE_REPLICATION_MIN_KEY,

19
                                        DFS_NAMENODE_REPLICATION_MIN_DEFAULT);

20
       
21
      LOG.info(DFS_NAMENODE_SAFEMODE_THRESHOLD_PCT_KEY + " = " + threshold);

22
      LOG.info(DFS_NAMENODE_SAFEMODE_MIN_DATANODES_KEY+"="+datanodeThreshold);

23
      LOG.info(DFS_NAMENODE_SAFEMODE_EXTENSION_KEY + "     = " + extension);

24
 
25
      this.replQueueThreshold =

26
        conf.getFloat(DFS_NAMENODE_REPL_QUEUE_THRESHOLD_PCT_KEY,

27
                      (float) threshold);

28
      this.blockTotal = 0;

29
      this.blockSafe = 0;

30
}

上述代码从配置文件中获取dfs.namenode.safemode.min.datanodes、dfs.namenode.safemode.threshold-pct、dfs.namenode.replication.min和dfs.namenode.safemode.extension。　　我们需要知道，当Resource is low时，只有Active NameNode才会进入到安全模式（HDFS-2914）。　　当然，如果我们也可以手动进入安全模式