CloudStack VM运行状态的监控-Hypervisor
接上篇：[CloudStack VM运行状态的监控-Management]
本文继续讲解Hyperviser端的处理逻辑.
XenServer
XenServer的处理逻辑在CitrixResourceBase.java 中，该类封装了几乎全部的XenServer操作，主要调用XenServer API 和 XenServer RRD API来完成。
@Override
public Answer executeRequest(Command cmd) {
Class<? extends Command> clazz = cmd.getClass();
if (clazz == CreateCommand.class) {
......
} else if (clazz == GetVmStatsCommand.class) {
return execute((GetVmStatsCommand)cmd);
}
......
}
该方法判断接收到的Command类型，并dispatch到相关处理方法。
protected GetVmStatsAnswer execute(GetVmStatsCommand cmd) {
......
HashMap<String, VmStatsEntry> vmStatsUUIDMap = getVmStats(conn, cmd, vmUUIDs, cmd.getHostGuid());
......
for (String vmUUID : vmStatsUUIDMap.keySet()) {
vmStatsNameMap.put(vmNames.get(vmUUIDs.indexOf(vmUUID)), vmStatsUUIDMap.get(vmUUID));
}
......
return new GetVmStatsAnswer(cmd, vmStatsNameMap);
}
该方法调用getVmStats()方法，获取数据，并进行处理后返回。
protected HashMap<String, VmStatsEntry> getVmStats(Connection conn, GetVmStatsCommand cmd, List<String> vmUUIDs, String hostGuid) {
HashMap<String, VmStatsEntry> vmResponseMap = new HashMap<String, VmStatsEntry>();
......
Object[] rrdData = getRRDData(conn, 2); // call rrddata with 2 for vm
......//解析RRD Data
if (param.contains("cpu")) {
vmStatsAnswer.setNumCPUs(vmStatsAnswer.getNumCPUs() + 1);
vmStatsAnswer.setCPUUtilization(((vmStatsAnswer.getCPUUtilization() + getDataAverage(dataNode, col, numRows))));
} else if (param.matches("vif_\\d*_rx")) {
vmStatsAnswer.setNetworkReadKBs(vmStatsAnswer.getNetworkReadKBs() + (getDataAverage(dataNode, col, numRows)/1000));
} else if (param.matches("vif_\\d*_tx")) {
vmStatsAnswer.setNetworkWriteKBs(vmStatsAnswer.getNetworkWriteKBs() + (getDataAverage(dataNode, col, numRows)/1000));
} else if (param.matches("vbd_.*_read")) {
vmStatsAnswer.setDiskReadKBs(vmStatsAnswer.getDiskReadKBs() + (getDataAverage(dataNode, col, numRows)/1000));
} else if (param.matches("vbd_.*_write")) {
vmStatsAnswer.setDiskWriteKBs(vmStatsAnswer.getDiskWriteKBs() + (getDataAverage(dataNode, col, numRows)/1000));
}
}
}
......
return vmResponseMap;
}
至此，全部逻辑处理完毕，XenServer 的 RRD 格式解析，日后会发文详解。 值得注意的是：getDataAverage()，看名字则可知其实逻辑是获取平均值。因为XenServer的RRD并不记录某段时间内的统计量，而是记录某一时刻的瞬时值。所以XenServer中，以此逻辑进行了处理，获取了该段时间内的全部瞬时值，然后进行处理得到近似这段时间内数据总量的统计值。
VMware
逻辑类似于XenServer，只看getVmStats() 方法
private HashMap<String, VmStatsEntry> getVmStats(List<String> vmNames) throws Exception {
......
List<PerfCounterInfo> cInfo = getServiceContext().getVimClient().getDynamicProperty(perfMgr, "perfCounter");
......
int key = ((HostMO)hyperHost).getCustomFieldKey("VirtualMachine", CustomFieldConstants.CLOUD_VM_INTERNAL_NAME);
......
String instanceNameCustomField = "value[" + key + "]";
ObjectContent[] ocs =
hyperHost.getVmPropertiesOnHyperHost(new String[] {"name", "summary.config.numCpu", "summary.quickStats.overallCpuUsage", instanceNameCustomField});
if (ocs != null && ocs.length > 0) {
for (ObjectContent oc : ocs) {
List<DynamicProperty> objProps = oc.getPropSet();
if (objProps != null) {
......
for (DynamicProperty objProp : objProps) {
if (objProp.getName().equals("name")) {
vmNameOnVcenter = objProp.getVal().toString();
} else if (objProp.getName().contains(instanceNameCustomField)) {
if (objProp.getVal() != null)
vmInternalCSName = ((CustomFieldStringValue)objProp.getVal()).getValue();
} else if (objProp.getName().equals("summary.config.numCpu")) {
numberCPUs = objProp.getVal().toString();
} else if (objProp.getName().equals("summary.quickStats.overallCpuUsage")) {
maxCpuUsage = objProp.getVal().toString();
}
}
new VirtualMachineMO(hyperHost.getContext(), oc.getObj());
......
ManagedObjectReference vmMor = hyperHost.findVmOnHyperHost(name).getMor();
......
if (vmNetworkMetrics.size() != 0) {
......
for (int i = 0; i < values.size(); ++i) {
List<PerfSampleInfo> infos = ((PerfEntityMetric)values.get(i)).getSampleInfo();
if (infos != null && infos.size() > 0) {
......
List<PerfMetricSeries> vals = ((PerfEntityMetric)values.get(i)).getValue();
for (int vi = 0; ((vals != null) && (vi < vals.size())); ++vi) {
if (vals.get(vi) instanceof PerfMetricIntSeries) {
PerfMetricIntSeries val = (PerfMetricIntSeries)vals.get(vi);
List<Long> perfValues = val.getValue();
Long sumRate = 0L;
for (int j = 0; j < infos.size(); j++) { // Size of the array matches the size as the PerfSampleInfo
sumRate += perfValues.get(j);
}
Long averageRate = sumRate / infos.size();
if (vals.get(vi).getId().getCounterId() == rxPerfCounterInfo.getKey()) {
networkReadKBs = sampleDuration * averageRate; //get the average RX rate multiplied by sampled duration
}
if (vals.get(vi).getId().getCounterId() == txPerfCounterInfo.getKey()) {
networkWriteKBs = sampleDuration * averageRate;//get the average TX rate multiplied by sampled duration
}
......
return vmResponseMap;
}
由逻辑可见，Vmware也是获取平均值，然后乘相关时间，得到近似这段时间内数据总量的统计值。
KVM
逻辑类似于XenServer，只看getVmStats() 方法
VmStatsEntry getVmStat(Connect conn, String vmName) throws LibvirtException {
......
if (oldStats != null) {
elapsedTime = now.getTimeInMillis() - oldStats._timestamp.getTimeInMillis();
double utilization = (info.cpuTime - oldStats._usedTime) / ((double)elapsedTime * 1000000);
utilization = utilization / node.cpus; //获取CPU使用值的平均值
......
}
.....。
for (InterfaceDef vif : vifs) {
DomainInterfaceStats ifStats = dm.interfaceStats(vif.getDevName());
rx += ifStats.rx_bytes;
tx += ifStats.tx_bytes;
}
if (oldStats != null) {
double deltarx = rx - oldStats._rx;
......
double deltatx = tx - oldStats._tx;
......
}
/* get disk stats */
......
for (DiskDef disk : disks) {
DomainBlockStats blockStats = dm.blockStats(disk.getDiskLabel());
io_rd += blockStats.rd_req;
io_wr += blockStats.wr_req;
bytes_rd += blockStats.rd_bytes;
bytes_wr += blockStats.wr_bytes;
}
if (oldStats != null) {
long deltaiord = io_rd - oldStats._ioRead;
......
long deltaiowr = io_wr - oldStats._ioWrote;
......
double deltabytesrd = bytes_rd - oldStats._bytesRead;
......
double deltabyteswr = bytes_wr - oldStats._bytesWrote;
......
}
......
return stats;
......
}
看代码逻辑，则可以得知，KVM得到的是准确的当前时间段内的磁盘和网络IO量。