oozie概述：

oozie是基于hadoop的调度器，以xml的形式写调度流程，可以调度mr，pig，hive，shell，jar等等。

主要的功能有

Workflow： 顺序执行流程节点，支持fork（分支多个节点），join（合并多个节点为一个）

Coordinator，定时触发workflow

Bundle Job，绑定多个coordinator

oozie格式：

写一个oozie，有两个是必要的：job.properties 和 workflow.xml(coordinator.xml,bundle.xml)

一、job.properties里定义环境变量

注意：

workflow：oozie.wf.application.path

coordinator：oozie.coord.application.path

bundle：oozie.bundle.application.path

二、XML

 1.workflow：

Xml代码  

1. <workflow-app xmlns="uri:oozie:workflow:0.2" name="wf-example1">  
2.   <start to="pig-node">  
3.   <action name="pig-node">  
4.       <pig>  
5.            <job-tracker>${jobTracker}job-tracker>  
6.            <name-node>${nameNode}name-node>  
7.            <prepare>  
8.               <delete path="hdfs://xxx5/user/hadoop/appresult" />  
9.            prepare>  
10.            <configuration>  
11.              <property>  
12.                   <name>mapred.job.queue.namename>  
13.                    <value>defaultvalue>  
14.               <property>  
15.              <property>  
16.                   <name>mapred.compress.map.outputname>  
17.                    <value>truevalue>  
18.               <property>  
19.              <property>  
20.                   <name>mapreduce.fileoutputcommitter.marksuccessfuljobsname>  
21.                    <value>falsevalue>  
22.               <property>  
23.            configuration>  
24.             <script>test.pigscript>  
25.             <param>filepath=${filpath}param>  
26.       pig>  
27.        <ok to="end">  
28.         <error to="fail">  
29.   action>  
30.   <kill name="fail">  
31.        <message>  
32.                    Map/Reduce failed, error               message[${wf:errorMessage(wf:lastErrorNode())}]  
33.          message>  
34.     kill>  
35.    <end name="end"/>  
36. workflow-app>  

  
  
      
           ${jobTracker}
           ${nameNode}
           
              
           
           
             
                  mapred.job.queue.name
                   default
              
             
                  mapred.compress.map.output
                   true
              
             
                  mapreduce.fileoutputcommitter.marksuccessfuljobs
                   false
              
           
            
            filepath=${filpath}
      
       
        
  
  
       
                   Map/Reduce failed, error               message[${wf:errorMessage(wf:lastErrorNode())}]
         
    
   

2.coordinator

Xml代码  

1. <coordinator-app name="cron-coord" frequence="${coord:hours(6)}" start="${start}" end="${end}"  
2. timezoe="UTC" xmlns="uri:oozie:coordinator:0.2">  
3. <action>  
4. <workflow>  
5.    <app-path>${nameNode}/user/{$coord:user()}/${examplesRoot}/wpathapp-path>  
6.   <configuration>  
7.            <property>  
8.            <name>jobTrackername>  
9.            <value>${jobTracker}value>  
10.            property>  
11.            <property>  
12.            <name>nameNodename>  
13.            <value>${nameNode}value>  
14.            property>  
15.            <property>  
16.            <name>queueNamename>  
17.            <value>${queueName}value>  
18.        property>  
19.   configuration>  
20. workflow>  
21. action>  

   ${nameNode}/user/{$coord:user()}/${examplesRoot}/wpath
  
           
           jobTracker
           ${jobTracker}
           
           
           nameNode
           ${nameNode}
           
           
           queueName
           ${queueName}
       
  

 注意：coordinator设置的UTC，比北京时间晚8个小时，所以你要是把期望执行时间减8小时

3.bundle

Java代码  

1. 'APPNAME' xmlns:xsi='http://www.w3.org/2001/XMLSchema-instance' xmlns='uri:oozie:bundle:0.1'>   
2.     
3.        ${kickOffTime}  
4.     
5.    'coordJobFromBundle1' >  
6.        ${appPath}  
7.          
8.            
9.               startTime1  
10.               ${START_TIME}  
11.             
12.            
13.               endTime1  
14.               ${END_TIME}  
15.             
16.         
17.      
18.    'coordJobFromBundle2' >  
19.        ${appPath2}  
20.          
21.            
22.               startTime2  
23.               ${START_TIME2}  
24.             
25.            
26.               endTime2  
27.               ${END_TIME2}  
28.             
29.         
30.      
31.   

 
  
       ${kickOffTime}
  
   
       ${appPath}
       
         
              startTime1
              ${START_TIME}
          
         
              endTime1
              ${END_TIME}
          
      
   
   
       ${appPath2}
       
         
              startTime2
              ${START_TIME2}
          
         
              endTime2
              ${END_TIME2}
          
      
   

oozie运行

启动任务：

Java代码  

1. oozie job -oozie http://xxx5:11000/oozie -config job.properties -run  

oozie job -oozie http://xxx5:11000/oozie -config job.properties -run

停止任务：

oozie job -oozie http://localhost:8080/oozie -kill 14-20090525161321-oozie-joe

注意：在停止任务的时候，有的时候会出现全线问题，需要修改oozie-site.xml文件

hadoop.proxyuser.oozie.groups *

hadoop.proxyuser.oozie.hosts *

oozie.server.ProxyUserServer.proxyuser.hadoop.hosts *

oozie.server.ProxyUserServer.proxyuser.hadoop.groups *

原文：http://blackproof.iteye.com/blog/1928122