Hadoop自定义JobTracker和NameNode管理页面

为了可以方便地在hadoop的管理界面（namenode和jobtracker）中自定义展示项，使用代理servlet的方式实现了hadoop的管理界面。

首先，

在org.apache.hadoop.http.HttpServer中的构造函数public HttpServer(String name, String bindAddress, int port,boolean findPort, Configuration conf, AccessControlList adminsAcl,Connector connector)中添加如下代码，指定资源包和url前缀：

ServletHolder servletHoder = new ServletHolder();
servletHoder.setInitParameter("com.sun.jersey.config.property.packages","com.test.agent");
servletHoder.setServlet(new com.sun.jersey.spi.container.servlet.ServletContainer());
webAppContext.addServlet(servletHoder, "/agent/*");

之后，在包中定义了一个抽象对象AbstractResource　

public class AbstractResource {
	protected JsonMapper jsonMapper = JsonMapper.nonDefaultMapper();
	protected @Context
	ServletContext context;
	public JsonMapper getJsonMapper() {
		return jsonMapper;
	}
	public void setJsonMapper(JsonMapper jsonMapper) {
		this.jsonMapper = jsonMapper;
	}
	public ServletContext getContext() {
		return context;
	}
	public void setContext(ServletContext context) {
		this.context = context;
	}
}

然后，真实的资源代码实现AbstractResource对象，对资源进行定义

@Path("/job")
public class JobResource extends AbstractResource {

	private DecimalFormat percentFormat = new DecimalFormat("##0.00");
	private SimpleDateFormat dateFormat = new SimpleDateFormat(
			"d-MMM-yyyy HH:mm:ss");

	@Path("/clustersummary")
	@GET
	@Consumes(MediaType.TEXT_PLAIN)
	@Produces(MediaType.TEXT_PLAIN)
	public String getClusterSummaryInfo() {
		JobTracker jt = (JobTracker) context.getAttribute("job.tracker");
		ClusterMetrics metrics = jt.getClusterMetrics();
		String tasksPerNode = metrics.getTaskTrackerCount() > 0 ? percentFormat
				.format(((double) (metrics.getMapSlotCapacity() + metrics
						.getReduceSlotCapacity()))
						/ metrics.getTaskTrackerCount()) : "-";
		Map<String, String> summaryInfo = new LinkedHashMap<String, String>();
		summaryInfo.put("usedHeapMemoryBytes",
				Long.toString(Runtime.getRuntime().totalMemory()));
		summaryInfo.put("totalHeapMemoryBytes",
				Long.toString(Runtime.getRuntime().maxMemory()));
		summaryInfo.put("runningMapTasks",
				Integer.toString(metrics.getRunningMaps()));
		summaryInfo.put("runningReduceTasks",
				Integer.toString(metrics.getRunningReduces()));
		summaryInfo.put("totalJobSubmissions",
				Integer.toString(metrics.getTotalJobSubmissions()));
		summaryInfo.put("numTotalTaskTrackers",
				Integer.toString(metrics.getTaskTrackerCount()));
		summaryInfo.put("occupiedMapSlots",
				Integer.toString(metrics.getOccupiedMapSlots()));
		summaryInfo.put("occupiedReduceSlots",
				Integer.toString(metrics.getOccupiedReduceSlots()));
		summaryInfo.put("reservedMapSlots",
				Integer.toString(metrics.getReservedMapSlots()));
		summaryInfo.put("reservedReduceSlots",
				Integer.toString(metrics.getReservedReduceSlots()));
		summaryInfo.put("mapTaskCapacity",
				Integer.toString(metrics.getMapSlotCapacity()));
		summaryInfo.put("reduceTaskCapacity",
				Integer.toString(metrics.getReduceSlotCapacity()));
		summaryInfo.put("avgTasksPerTaskTracker", tasksPerNode);
		summaryInfo.put("numBlackListedTaskTrackers",
				Integer.toString(metrics.getBlackListedTaskTrackerCount()));
		summaryInfo.put("numGrayListedTaskTrackers",
				Integer.toString(metrics.getGrayListedTaskTrackerCount()));
		summaryInfo.put("numDecommissionedTaskTrackers",
				Integer.toString(metrics.getDecommissionedTaskTrackerCount()));
		return jsonMapper.toJson(summaryInfo);

	}
}

这样可以通过访问url进行访问，自定义管理界面访问的端口与原有的管理页面端口相同。JobTracker使用50030，NameNode使用50070，依次类推。具体见下单元测试代码：

public class JobResourceIntTest {
	HttpClient client;

	@Before
	public void setUp() {
		client = new DefaultHttpClient();
	}

	private String urlPrefix = "http://ip:50030/agent/job/";

	private String jobId = "job_201405121106_0001";
	private String tipId = "task_201405121106_0001_m_000000";
	private String attemptId = "attempt_201405121106_0001_m_000000_0";

	private String regex = "\"total\":\"(.+?)\",";
	private Pattern p = Pattern.compile(regex);

	@Test
	public void testClusterSummary() {
		HttpGet clusterSummary = new HttpGet(urlPrefix + "clustersummary");
		try {
			HttpResponse response = client.execute(clusterSummary);
			String clusterSummaryStr = EntityUtils.toString(response
					.getEntity());
			System.out.println(clusterSummaryStr);
			Assert.assertTrue(clusterSummaryStr.contains("usedHeapMemoryBytes"));
			Assert.assertTrue(clusterSummaryStr.contains("runningMapTasks"));
		} catch (Exception e) {
			e.printStackTrace();
		}
		clusterSummary.releaseConnection();
	}
	@After
	public void tearDown() {
	}
}

上述是后端的实现和单元测试，前端可根据需要进行资源获取和界面设计。