wiki:Hadoop_Lab8_018

Version 1 (modified by waue, 15 years ago) (diff)

--

實作八: Hadoop 叢集進階用法

狀況一: 如何動態加入datanode 與 tasktracker

  • 某些情況下,後來的環境也許會跟之前配置的不同,比如說,原本只有五台機器架設hadoop,但是也許老闆今天心血來潮,又撥了五台電腦給你。在接續之前的環境動態的擴增節點的方法,請看以下作法。

1.0 說明

  • 要新增的節點上,hadoop版本與設定檔要與原先的相同
  • 是否能連到正確的位址取決於conf/hadoop-site.xml內的jobTracker, Namenode資訊是否正確 (目前測試結果與conf/slave、masters無關)

1.1 加入datanode

  • 在要加入的節點上面下此指令
    $ cd $HADOOP_HOME
    $ bin/hadoop-daemon.sh --config ./conf start datanode
    
    • 執行畫面如下:
      starting datanode, logging to /tmp/hadoop/logs/hadoop-waue-datanode-Dx7200.out
      

1.2 加入 taskTracker

  • 是否能連到正確的namenode取決於conf/hadoop-site.xml,目前測試結果與conf/slave、masters無關
    $ cd $HADOOP_HOME
    $ bin/hadoop-daemon.sh --config ./conf start tasktracker
    
    • 執行畫面如下:
      starting tasktracker, logging to /tmp/hadoop/logs/hadoop-waue-tasktracker-Dx7200.out
      

狀況二: 怎麼讓我的HDFS內資料均勻分散於各個Node上

  • 下面指令用於分析數據塊分佈和重新平衡!DataNode上的數據分佈
    $ bin/hadoop balancer
    
    • 執行畫面如下
      Time Stamp               Iteration#  Bytes Already Moved  Bytes Left To Move  Bytes Being Moved
      09/04/01 18:00:08 INFO net.NetworkTopology: Adding a new node: /default-rack/140.110.138.191:50010
      09/04/01 18:00:08 INFO net.NetworkTopology: Adding a new node: /default-rack/140.110.141.129:50010
      09/04/01 18:00:08 INFO dfs.Balancer: 0 over utilized nodes:
      09/04/01 18:00:08 INFO dfs.Balancer: 0 under utilized nodes: 
      The cluster is balanced. Exiting...
      Balancing took 186.0 milliseconds
      

狀況三:如何讓已上線服務的Hadoop進行升級並且不失去以前的資料

  • 假設從原本的hadoop 0.16升級到hadoop 0.18
  • 如果把conf/這個資料夾至於$Hadoop_home目錄下的話,一旦換版本就連conf也被一併換掉,但無論hadoop的版本新舊,其實設定檔及其資訊是可以共用的。

step 1. 停止hdfs

  • 先看狀態
    $ cd /opt/hadoop/
    $ bin/hadoop dfsadmin -upgradeProgress status
    
    There are no upgrades in progress.
    
  • 停止hdfs
    • 注意不可使用bin/stop-all.sh來停止
      $ bin/stop-dfs.sh
      

Step 2. 鍊結新版本hadoop

  • 把conf 移至/opt/conf ,hadoop 0.16 與 hadoop 0.18用 ln 做捷徑代換。
  • 以下假設你已經下載好hadoop0.18並解壓縮後,資料夾名稱為hadoop-0.18.3
    $ cd opt/
    $ mv hadoop/conf ./
    $ mv hadoop hadoop-0.16
    $ ln hadoop-0.18.3 hadoop
    

step 3. 設置環境變數

  • 由於conf已不在hadoop_home內,因此記得匯入conf/hadoop-env.sh的參數
  • 填入hadoop-env.sh 內$HADOOP_CONF_DIR正確路徑,並匯入資訊
    $ source /opt/conf/hadoop-env.sh
    

step 4. 每個節點都部署新版本的Hadoop

  • 若有多個node的話,則每個node的hadoop版本都要統一,否則會出現問題

step 5. 啟動

$ bin/start-dfs.sh -upgrade
  • namenode管理網頁會出現升級狀態

狀況四:如何讓已上線服務的Hadoop進行降級並且不失去以前的資料

  • 此情況與狀況三相反,因此作法類似狀況三,下面的狀況假設設定檔已在/opt/conf內,並且/opt內也都有hadoop-0.16 與 hadoop-0.18.3 兩個資料夾,而且節點只有一台。

step 1. 停止 HDFS

$ cd /opt/hadoop
$ bin/stop-dfs.sh

step 2. 部署老版本的Hadoop

$ rm /opt/hadoop
$ ln hadoop-0.16 hadoop

step 3. 退回之前版本

$ bin/start-dfs.sh -rollback

狀況五:我的HDFS檔案系統是否正常

  • 在此介紹HDFS文件系統檢查工具 "bin/hadoop fsck"
$ bin/hadoop fsck /
  • 執行畫面如下
    .
    /user/waue/input/1.txt:  Under replicated blk_-90085106852013388_1001. Target Replicas is 3 but found 2 replica(s).
    /user/waue/input/1.txt:  Under replicated blk_-4027196261436469955_1001. Target Replicas is 3 but found 2 replica(s).
    .
    /user/waue/input/2.txt:  Under replicated blk_-2300843106107816641_1002. Target Replicas is 3 but found 2 replica(s).
    .
    /user/waue/input/3.txt:  Under replicated blk_-1561577350198661966_1003. Target Replicas is 3 but found 2 replica(s).
    .
    /user/waue/input/4.txt:  Under replicated blk_1316726598778579026_1004. Target Replicas is 3 but found 2 replica(s).
    Status: HEALTHY
     Total size:	143451003 B
     Total dirs:	8
     Total files:	4
     Total blocks (validated):	5 (avg. block size 28690200 B)
     Minimally replicated blocks:	5 (100.0 %)
     Over-replicated blocks:	0 (0.0 %)
     Under-replicated blocks:	5 (100.0 %)
     Mis-replicated blocks:		0 (0.0 %)
     Default replication factor:	3
     Average block replication:	2.0
     Corrupt blocks:		0
     Missing replicas:		5 (50.0 %)
     Number of data-nodes:		2
     Number of racks:		1
    The filesystem under path '/' is HEALTHY
    
  • 加不同的參數有不同的用處,如
    $ bin/hadoop fsck / -files
    
    • 執行畫面如下
      /tmp <dir>
      /tmp/hadoop <dir>
      /tmp/hadoop/hadoop-waue <dir>
      /tmp/hadoop/hadoop-waue/mapred <dir>
      /tmp/hadoop/hadoop-waue/mapred/system <dir>
      /user <dir>
      /user/waue <dir>
      /user/waue/input <dir>
      /user/waue/input/1.txt 115045564 bytes, 2 block(s):  Under replicated blk_-90085106852013388_1001. Target Replicas is 3 but found 2 replica(s).
       Under replicated blk_-4027196261436469955_1001. Target Replicas is 3 but found 2 replica(s).
      /user/waue/input/2.txt 987864 bytes, 1 block(s):  Under replicated blk_-2300843106107816641_1002. Target Replicas is 3 but found 2 replica(s).
      /user/waue/input/3.txt 1573048 bytes, 1 block(s):  Under replicated blk_-1561577350198661966_1003. Target Replicas is 3 but found 2 replica(s).
      /user/waue/input/4.txt 25844527 bytes, 1 block(s):  Under replicated blk_1316726598778579026_1004. Target Replicas is 3 but found 2 replica(s).
      Status: HEALTHY
      ....(同上)
      

狀況六:我的系統似乎跑太多Job,我要幫他減肥

step 1. 把所有程序列出來

  • 可到JobTracker:50030網頁來看程序的Jobid
  • 或用指令印出所有程序
    $ bin/hadoop job -list all
    
    5 jobs submitted
    States are:
    	Running : 1	Succeded : 2	Failed : 3	Prep : 4
    JobId	State	StartTime	UserName
    job_200904021140_0001	2	1238652150499	waue
    job_200904021140_0002	3	1238657754096	waue
    job_200904021140_0004	3	1238657989495	waue
    job_200904021140_0005	2	1238658076347	waue
    job_200904021140_0006	2	1238658644666	waue
    

step 2. more detail

  • 查看工作狀態
    $ bin/hadoop job -status job_200904021140_0001
    
  • 印出程序的歷史狀態
    $ bin/hadoop job -history /user/waue/stream-output1
    
    Hadoop job: job_200904021140_0005
    =====================================
    Job tracker host name: gm1.nchc.org.tw
    job tracker start time: Thu Apr 02 11:40:06 CST 2009
    User: waue
    JobName: streamjob9019.jar
    JobConf: hdfs://gm1.nchc.org.tw:9000/tmp/hadoop/hadoop-waue/mapred/system/job_200904021140_0005/job.xml
    Submitted At: 2-四月-2009 15:41:16
    Launched At: 2-四月-2009 15:41:16 (0sec)
    Finished At: 2-四月-2009 15:42:04 (48sec)
    Status: SUCCESS
    =====================================
    ...略
    

step 3. 殺很大、殺不用錢

  • 終止正在執行的程序,如 id 為 job_200904021140_0001
    $ bin/hadoop job -kill job_200904021140_0001
    

狀況七:怎麼看目前的Hadoop版本

  • 印出目前的hadoop 版本
    $ bin/hadoop version
    
    • 執行畫面
      Hadoop 0.18.3
      Subversion https://svn.apache.org/repos/asf/hadoop/core/branches/branch-0.18 -r 736250
      Compiled by ndaley on Thu Jan 22 23:12:08 UTC 2009
      

狀況八:我要設定HDFS的帳戶及配額

step 1. 先設定各使用者的預設資料夾,屬性及讀寫權限

  • hdfs的權限有owner, group, other三種
  • 而用戶的身份取決於client上的使用者 (用 whoami),群組為(bash -c groups)
  • 相關的操作:
    $ bin/hadoop fs -mkdir own
    $ bin/hadoop fs -chmod -R 755 own
    $ bin/hadoop fs -chgrp -R waue own
    $ bin/hadoop fs -chown -R waue own
    $ bin/hadoop fs -lsr own
    
  • conf/hadoop-site.xml 可用參數:
    dfs.permissions = true 
    dfs.web.ugi = webuser,webgroup
    dfs.permissions.supergroup = supergroup
    dfs.upgrade.permission = 777
    dfs.umask = 022
    

step 2. 設定配額

  • 目錄配額是對目錄樹上該目錄下的名字數量做硬性限制
  • 設定配額,數字代表個數 (如:我上傳了一個2個block的檔案可以上傳,但我上傳兩個檔案很小的檔上去卻不行)
  • 配額為1可以強制目錄保持為空
  • 重命名不會改變該目錄的配額
    $ bin/hadoop fs -mkdir quota
    $ bin/hadoop dfsadmin -setQuota 2 quota
    $ bin/hadoop fs -put ../conf/hadoop-env.sh quota/
    $ bin/hadoop fs -put ../conf/hadoop-site.xml quota/
    
    put: org.apache.hadoop.dfs.QuotaExceededException: The quota of /user/waue/quota is exceeded: quota=2 count=3
    
  • 檢查目錄的配額方法: "bin/hadoop fs -count -q <目錄> "
$ bin/hadoop fs -count -q own
   none  inf  1    0     0 hdfs://gm1.nchc.org.tw:9000/user/waue/own
$ bin/hadoop dfsadmin -setQuota 4 own
$ bin/hadoop fs -count -q own
    4     3   1    0     0 hdfs://gm1.nchc.org.tw:9000/user/waue/own
  • 清除之前設定的配額
    $ bin/hadoop dfsadmin -clrQuota quota/