你懂集群monitoring么？（二）—— HDFS部分指标

标签：sum   local   cal   过程   ISE   getbean   map   ini   ssi   

本篇文章接着上篇内容继续，地址：IDC集群相关指标获取
在获取了对应的IDC机器自身的指标之后，还需要对Hadoop集群中HDFS和YARN的指标进行采集，大体思路上可以有2种：

第一种当然还是可以延用CM API去获取，因为CM中的tssql提供了非常丰富的各种指标监控
第二种即通过jmxJ去获取数据，其实就是通过访问上述这些相关的URL，然后将得到的json进行解析，从而获取到我们需要的数据，最终将这些数据归并到一起，定时的去执行采集操作
在实际的实践过程当中使用jmx这种方式去进行获取，涉及到的url请求如下：
http://localhost:50070/jmx?qry=Hadoop:service=NameNode,name=NameNodeInfo
http://localhost:50070/jmx?qry=Hadoop:service=NameNode,name=FSNamesystemState

具体的代码实现思路如下：

首先需要有个httpclient，去向server发起请求，从而获得对应的json数据，这里自己编写了StatefulHttpClient
其次使用JsonUtil该工具类，用于Json类型的数据与对象之间的转换
当然，我们也需要将所需要获取的监控指标给梳理出来，编写我们的entity，这里以HDFS为例，主要为HdfsSummary和DataNodeInfo
本案例的代码在github上，地址：
这里主要展示核心的代码：

MonitorMetrics.java：

public class MonitorMetrics {
    // beans为通过jmx所返回的json串中最起始的key
    // 结构为{"beans":[{"":"","":"",...}]}
    List<Map<String, Object>> beans = new ArrayList<>();

    public List<Map<String, Object>> getBeans() {
        return beans;
    }

    public void setBeans(List<Map<String, Object>> beans) {
        this.beans = beans;
    }

    public Object getMetricsValue(String name) {
        if (beans.isEmpty()) {
            return null;
        }
        return beans.get(0).getOrDefault(name, null);
    }
}

HadoopUtil.java：

public class HadoopUtil {
    public static long gbLength = 1073741824L;
    public static final String hadoopJmxServerUrl = "http://localhost:50070";
    public static final String jmxServerUrlFormat = "%s/jmx?qry=%s";
    public static final String nameNodeInfo = "Hadoop:service=NameNode,name=NameNodeInfo";
    public static final String fsNameSystemState = "Hadoop:service=NameNode,name=FSNamesystemState";

    public static HdfsSummary getHdfsSummary(StatefulHttpClient client) throws IOException {
        HdfsSummary hdfsSummary = new HdfsSummary();
        String namenodeUrl = String.format(jmxServerUrlFormat, hadoopJmxServerUrl, nameNodeInfo);
        MonitorMetrics monitorMetrics = client.get(MonitorMetrics.class, namenodeUrl, null, null);
        hdfsSummary.setTotal(doubleFormat(monitorMetrics.getMetricsValue("Total"), gbLength));
        hdfsSummary.setDfsFree(doubleFormat(monitorMetrics.getMetricsValue("Free"), gbLength));
        hdfsSummary.setDfsUsed(doubleFormat(monitorMetrics.getMetricsValue("Used"), gbLength));
        hdfsSummary.setPercentUsed(doubleFormat(monitorMetrics.getMetricsValue("PercentUsed")));
        hdfsSummary.setSafeMode(monitorMetrics.getMetricsValue("Safemode").toString());
        hdfsSummary.setNonDfsUsed(doubleFormat(monitorMetrics.getMetricsValue("NonDfsUsedSpace"), gbLength));
        hdfsSummary.setBlockPoolUsedSpace(doubleFormat(monitorMetrics.getMetricsValue("BlockPoolUsedSpace"), gbLength));
        hdfsSummary.setPercentBlockPoolUsed(doubleFormat(monitorMetrics.getMetricsValue("PercentBlockPoolUsed")));
        hdfsSummary.setPercentRemaining(doubleFormat(monitorMetrics.getMetricsValue("PercentRemaining")));
        hdfsSummary.setTotalBlocks((int) monitorMetrics.getMetricsValue("TotalBlocks"));
        hdfsSummary.setTotalFiles((int) monitorMetrics.getMetricsValue("TotalFiles"));
        hdfsSummary.setMissingBlocks((int) monitorMetrics.getMetricsValue("NumberOfMissingBlocks"));

        String liveNodesJson = monitorMetrics.getMetricsValue("LiveNodes").toString();
        String deadNodesJson = monitorMetrics.getMetricsValue("DeadNodes").toString();
        List<DataNodeInfo> liveNodes = dataNodeInfoReader(liveNodesJson);
        List<DataNodeInfo> deadNodes = dataNodeInfoReader(deadNodesJson);
        hdfsSummary.setLiveDataNodeInfos(liveNodes);
        hdfsSummary.setDeadDataNodeInfos(deadNodes);

        String fsNameSystemStateUrl = String.format(jmxServerUrlFormat, hadoopJmxServerUrl, fsNameSystemState);
        MonitorMetrics hadoopMetrics = client.get(MonitorMetrics.class, fsNameSystemStateUrl, null, null);
        hdfsSummary.setNumLiveDataNodes((int) hadoopMetrics.getMetricsValue("NumLiveDataNodes"));
        hdfsSummary.setNumDeadDataNodes((int) hadoopMetrics.getMetricsValue("NumDeadDataNodes"));
        hdfsSummary.setVolumeFailuresTotal((int) hadoopMetrics.getMetricsValue("VolumeFailuresTotal"));

        return hdfsSummary;
    }

    public static List<DataNodeInfo> dataNodeInfoReader(String jsonData) throws IOException {
        List<DataNodeInfo> dataNodeInfos = new ArrayList<DataNodeInfo>();
        Map<String, Object> nodes = JsonUtil.fromJsonMap(String.class, Object.class, jsonData);
        for (Map.Entry<String, Object> node : nodes.entrySet()) {
            Map<String, Object> info = (HashMap<String, Object>) node.getValue();
            String nodeName = node.getKey().split(":")[0];
            DataNodeInfo dataNodeInfo = new DataNodeInfo();
            dataNodeInfo.setNodeName(nodeName);
            dataNodeInfo.setNodeAddr(info.get("infoAddr").toString().split(":")[0]);
            dataNodeInfo.setLastContact((int) info.get("lastContact"));
            dataNodeInfo.setUsedSpace(doubleFormat(info.get("usedSpace"), gbLength));
            dataNodeInfo.setAdminState(info.get("adminState").toString());
            dataNodeInfo.setNonDfsUsedSpace(doubleFormat(info.get("nonDfsUsedSpace"), gbLength));
            dataNodeInfo.setCapacity(doubleFormat(info.get("capacity"), gbLength));
            dataNodeInfo.setNumBlocks((int) info.get("numBlocks"));
            dataNodeInfo.setRemaining(doubleFormat(info.get("remaining"), gbLength));
            dataNodeInfo.setBlockPoolUsed(doubleFormat(info.get("blockPoolUsed"), gbLength));
            dataNodeInfo.setBlockPoolUsedPerent(doubleFormat(info.get("blockPoolUsedPercent")));

            dataNodeInfos.add(dataNodeInfo);
        }

        return dataNodeInfos;
    }

    public static DecimalFormat df = new DecimalFormat("#.##");

    public static double doubleFormat(Object num, long unit) {
        double result = Double.parseDouble(String.valueOf(num)) / unit;
        return Double.parseDouble(df.format(result));
    }

    public static double doubleFormat(Object num) {
        double result = Double.parseDouble(String.valueOf(num));
        return Double.parseDouble(df.format(result));
    }

    public static void main(String[] args) {
        String res = String.format(jmxServerUrlFormat, hadoopJmxServerUrl, nameNodeInfo);
        System.out.println(res);
    }

}

MonitorApp.java：

public class MonitorApp {

    public static void main(String[] args) throws IOException {
        StatefulHttpClient client = new StatefulHttpClient(null);
        HadoopUtil.getHdfsSummary(client).printInfo();

    }
}

最终展示结果如下：

关于YARN指标的获取，思路类似，这里就不再展示了

你懂集群monitoring么？（二）—— HDFS部分指标

标签：sum   local   cal   过程   ISE   getbean   map   ini   ssi   

原文地址：https://blog.51cto.com/14309075/2415632