这篇博客会介绍如何通过JMX 来获取JAVA 进程占用的CPU利用率和GC所占用的CPU利用率

---

在使用JVisualVM的时候,发现它可以查看当前JAVA 进程占用的CPU利用率和GC 所占用的CPU利用率,很奇怪它是如何计算的或者怎么获取的.
本文会根据JVisualVM的源码来描述JVisualVM是如何计算的.

获取源码

GITHUB 地址: https://github.com/visualvm/visualvm.src
百度云:http://pan.baidu.com/s/1boHI4HL

目标数据

运行jvisualvm, 目标是获取图中的CPU Usage 和 GC activity.

HOW TO

将源码导入Intellij后,可以全局搜索GC activity,然后可以查看到:
在Bundle.properties 中:

LBL_Gc_Usage=GC activity

然后查看这个property的使用地方:

com.sun.tools.visualvm.application.views.monitor.ApplicationMonitorView.CpuViewSupport
...

        private static final String CPU = NbBundle.getMessage(ApplicationMonitorView.class, "LBL_Cpu"); // NOI18N
        private static final String CPU_USAGE = NbBundle.getMessage(ApplicationMonitorView.class, "LBL_Cpu_Usage"); // NOI18N
        private static final String GC_USAGE = NbBundle.getMessage(ApplicationMonitorView.class, "LBL_Gc_Usage"); // NOI18N
...
private void initModels(ApplicationMonitorModel model) {
            liveModel = model.isLive();
            processorsCount = model.getProcessorsCount();
            cpuMonitoringSupported = model.isCpuMonitoringSupported();
            gcMonitoringSupported = model.isGcMonitoringSupported();

            SimpleXYChartDescriptor chartDescriptor =
                    //下面的这个后面会讲到
                    SimpleXYChartDescriptor.percent(false, 0.1d, model.getChartCache());

            chartDescriptor.addLineItems(CPU_USAGE, GC_USAGE);
            chartDescriptor.setDetailsItems(new String[] { CPU_USAGE, GC_USAGE });

            chartSupport = ChartFactory.createSimpleXYChart(chartDescriptor);
//这里就应该就是界面显示的注册了.   那么注册后是如何使用的???            model.registerCpuChartSupport(chartSupport);

            chartSupport.setZoomingEnabled(!liveModel);
        }

主要的计算逻辑集中在其refresh方法中:

public void refresh(ApplicationMonitorModel model) {
     //如果监控CPU
            if (cpuMonitoringSupported || gcMonitoringSupported) {
     //获取当前的UPTIME
                long upTime = model.getUpTime() * 1000000;
                //获取上次的UPTIME
                long prevUpTime = model.getPrevUpTime() * 1000000;

                boolean tracksProcessCpuTime = cpuMonitoringSupported &&
                                               model.getPrevProcessCpuTime() != -1;
                                               //获取当前的CPU时间
                long processCpuTime = tracksProcessCpuTime ?
                    model.getProcessCpuTime() / processorsCount : -1;
                    //获取上次的CPU时间
                long prevProcessCpuTime = tracksProcessCpuTime ?
                    model.getPrevProcessCpuTime() / processorsCount : -1;

                boolean tracksProcessGcTime  = gcMonitoringSupported &&
                                               model.getPrevProcessGcTime() != -1;
                                               //获取当前的GC占用时间
                long processGcTime  = tracksProcessGcTime  ?
                    model.getProcessGcTime() * 1000000 / processorsCount : -1;
                    //获取上次的GC占用时间
                long prevProcessGcTime  = tracksProcessGcTime  ?
                    model.getPrevProcessGcTime() * 1000000 / processorsCount : -1;

                if (prevUpTime != -1 && (tracksProcessCpuTime || tracksProcessGcTime)) {

                    long upTimeDiff = upTime - prevUpTime;
                    //分别计算CPU使用率和GC的CPU使用率.
                    long cpuUsage = -1;
                    long gcUsage = -1;
                    String cpuDetail = UNKNOWN;
                    String gcDetail = UNKNOWN;

                    if (tracksProcessCpuTime) {
                        long processTimeDiff = processCpuTime - prevProcessCpuTime;
                        cpuUsage = upTimeDiff > 0 ? Math.min((long)(1000 * (float)processTimeDiff /
                                                             (float)upTimeDiff), 1000) : 0;
                        cpuDetail = cpuUsage == -1 ? UNKNOWN : chartSupport.formatPercent(cpuUsage);
                    }

                    if (tracksProcessGcTime) {
                        long processGcTimeDiff = processGcTime - prevProcessGcTime;
                        gcUsage = upTimeDiff > 0 ? Math.min((long)(1000 * (float)processGcTimeDiff /
                                                            (float)upTimeDiff), 1000) : 0;
                        if (cpuUsage != -1 && cpuUsage < gcUsage) gcUsage = cpuUsage;
                        gcDetail = gcUsage == -1 ? UNKNOWN : chartSupport.formatPercent(gcUsage);
                    }

                    if (liveModel)
                        //设置视图的值 达到刷新的目的
                        //可以看到这个值和界面上显示是相差0.1倍的 原因是前面的那个chartDescriptor 有个0.1d的显示倍率.
                        chartSupport.addValues(model.getTimestamp(), new long[] { Math.max(cpuUsage, 0), Math.max(gcUsage, 0) });
                    chartSupport.updateDetails(new String[] { cpuDetail, gcDetail });

                }
            }
        }

可以看到逻辑主要是ApplicationMonitorModel model对象的相关方法中,下面介绍这些时间是如何获取的:
ApplicationMonitorModel的方法:
private void updateValues(final long time, final MonitoredData data)实现了简单的值更新.主要就是prevXXX = xxx

private void updateValues(final long time, final MonitoredData data) {
        timestamp = time;
        if (data != null) {
            prevUpTime = uptime;
            uptime = data.getUpTime();
            ....
            其他类似

具体实现就在:MonitoredData中.
getUpTime:

getUpTime获取upTime:
RuntimeMXBean.getUpTime
获取CPU时间:
/*
JMXPath _processCpuTimeJMXPath = new JMXPath("java.lang:type=OperatingSystem:ProcessCpuTime");
获取这个JMXPath的值就可以
*/

if (jmxSupport.hasProcessCPUTimeAttribute()) {
            processCpuTime = jmxSupport.getProcessCPUTime(); // 与
        }
获取GC时间: 累加的GC时间
 //   gcList = ManagementFactory.getGarbageCollectorMXBeans();        
        if (gcList != null && !gcList.isEmpty()) {
            for (GarbageCollectorMXBean gcBean : gcList) {
                collectionTime+=gcBean.getCollectionTime();
            }
        }