Hadoop收集作业执行状态信息

最近一个项目需要收集hadoop作业的执行状态的信息，我给出了以下的解决策略：

1、从Hadoop提供的jobtracker.jsp获取需要的信息，这里遇到的一个问题是里面使用了application作用域
JobTracker tracker = (JobTracker) application.getAttribute("job.tracker");
而Jetty服务器是嵌入到Hadoop的内部的，
org.apache.mapred.Jobtracker.java
InetSocketAddress infoSocAddr = NetUtils.createSocketAddr(
conf.get(JT_HTTP_ADDRESS, "0.0.0.0:50030"));
infoServer = new HttpServer("job", infoBindAddress, tmpInfoPort,
tmpInfoPort == 0, conf);
infoServer.setAttribute("job.tracker", this);
于是，如果想通过jsp页面获取统计信息的话，必须绕开Jetty服务器，或者在修改Jobtracker的中返回infoServer的一个引用，在代码中实现，不过显然这个需要修改Hadoop的核心代码，灵活性不高。
2、脚本解析jsp.
通过wget http://localhost:50030/jobtracker.jsp可以看到：

-----------------------------------------------------------------------------------------------
<b>State:</b> RUNNING<br>
<b>Started:</b> Tue Dec 28 09:43:40 CST 2010<br>
<b>Version:</b> 0.21.0,
985326<br>
<b>Compiled:</b> Tue Aug 17 01:02:28 EDT 2010 by
tomwhite from
branches/branch-0.21<br>
<b>Identifier:</b> 201012280943<br>
...............................................................................................
这些信息完全都可以使用python beautiful soup（http://www.crummy.com/software/BeautifulSoup/）来解析得到。

3、把你的hadooop版本升级到Hadoop-0.21.0，Cluster类

提供了丰富的API接口

	cancelDelegationToken(org.apache.hadoop.security.token.Token<org.apache.hadoop.mapreduce.security.token.delegation.DelegationTokenIdentifier> token) Cancel a delegation token from the JobTracker
void	close() Close the Cluster.
TaskTrackerInfo[]	getActiveTaskTrackers() Get all active trackers in the cluster.
Job[]	getAllJobs() Get all the jobs in cluster.
TaskTrackerInfo[]	getBlackListedTaskTrackers() Get blacklisted trackers.
QueueInfo[]	getChildQueues(String queueName) Returns immediate children of queueName.
ClusterMetrics	getClusterStatus() Get current cluster status.
org.apache.hadoop.security.token.Token <org.apache.hadoop.mapreduce. security.token.delegation.DelegationTokenIdentifier>	getDelegationToken(org.apache.hadoop.io.Text renewer) Get a delegation token for the user from the JobTracker.
org.apache.hadoop.fs.FileSystem	getFileSystem() Get the file system where job-specific files are stored
Job	getJob(JobID jobId) Get job corresponding to jobid.
String	getJobHistoryUrl(JobID jobId) Get the job history file path for a given job id.
State	getJobTrackerState() Get JobTracker's state
QueueInfo	getQueue(String name) Get queue information for the specified name.
QueueAclsInfo[]	getQueueAclsForCurrentUser() Gets the Queue ACLs for current user
QueueInfo[]	getQueues() Get all the queues in cluster.
QueueInfo[]	getRootQueues() Gets the root level queues.
org.apache.hadoop.fs.Path	getStagingAreaDir() Grab the jobtracker's view of the staging directory path where job-specific files will be placed.
org.apache.hadoop.fs.Path	getSystemDir() Grab the jobtracker system directory path where job-specific files will be placed.

例如我们需要打印作业的信息的时候，只需要：
Configuration conf = new Configuration();
Cluster cluster = new Cluster(conf);
Job[] job = cluster.getAllJobs();
if(job != null) {
for (Job tmp : job) {
System.out.println(tmp.getJobID());
System.out.println(tmp.getJobName());
System.out.println(tmp.getStartTime());
System.out.println(tmp.getFinishTime());
}
}
:-)...

Klose 我们应该和Hadoop一起进步。