使用ranger可以控制用户提交作业。在搭建yarn插件之后,建立新的service的时候,填写入‘YARN REST URL ’后,点击test connection,出现了问题,页面弹出框提示无法通过该url获得队列,查看ranger-admin运行日志也没有报错。翻看底层代码后发现,ranger在对yarn的支持上只支持了capacity调度,对fair调度没有支持。
当将运行中的集群的调度改为capacity调度后,test connection成功。根本原因就是,resourcemanager的rest api http://rm:port/ws/v1/cluster/scheduler在不同的调度器下,得到的json文件不同,由于ranger底层只解析了capacity调度的json文件,没有解析fair调度的,所以出现了上述问题。
在ranger页面中点击test connection之后,对应的底层代码在YarnClient这个类中,其中的getQueuelist()方法中,通过Gson类中的fromJson(string,class)方法,来将string类型的json字符串序列化为某个class类的实例,这样就可以通过面向对象的方式获得json字符中的各项值。fromJson方法得到的值赋给了YarnSchdulerResponse对象的实例,我们需要修改的就是这个类。在其中添加针对fair调度生成的json文件的解析。出问题的代码片段
Gson gson = new GsonBuilder().setPrettyPrinting().create();
YarnSchedulerResponse yarnQResponse = gson.fromJson(jsonString, YarnSchedulerResponse.class);
if (yarnQResponse != null) {
List<String> yarnQueueList = yarnQResponse.getQueueNames();
if (yarnQueueList != null) {
for ( String yarnQueueName : yarnQueueList) {
if ( existingQueueList != null && existingQueueList.contains(yarnQueueName)) {
continue;
}
if (queueNameMatching == null || queueNameMatching.isEmpty()
|| yarnQueueName.startsWith(queueNameMatching)) {
if (LOG.isDebugEnabled()
...... lret.add(yarnQueueName);
}
}
}
问题就处在yarnQueueList这个实例中,在fair调度下,这个List为空无任何值,也就是没有拿到任何队列。
原YarnSchedulerResponse代码就不贴了,贴上修改过后的整个类的代码,也很简单。
import java.util.ArrayList;
import java.util.List;
import javax.xml.bind.annotation.XmlAccessType;
import javax.xml.bind.annotation.XmlAccessorType;
import javax.xml.bind.annotation.XmlRootElement;
import org.codehaus.jackson.annotate.JsonAutoDetect;
import org.codehaus.jackson.annotate.JsonAutoDetect.Visibility;
import org.codehaus.jackson.annotate.JsonIgnoreProperties;
import org.codehaus.jackson.map.annotate.JsonSerialize;
public class YarnSchedulerResponse implements java.io.Serializable{
private static final long serialVersionUID = 1L;
private YarnScheduler scheduler = null;
public YarnScheduler getScheduler() { return scheduler; }
public List<String> getQueueNames() {
List<String> ret = new ArrayList<String>();
if(scheduler != null) {
scheduler.collectQueueNames(ret);
}
return ret;
}
@JsonAutoDetect(getterVisibility=Visibility.NONE, setterVisibility=Visibility.NONE, fieldVisibility=Visibility.ANY)
@JsonSerialize(include=JsonSerialize.Inclusion.NON_NULL )
@JsonIgnoreProperties(ignoreUnknown=true)
@XmlRootElement
@XmlAccessorType(XmlAccessType.FIELD)
public static class YarnScheduler implements java.io.Serializable {
private static final long serialVersionUID = 1L;
private YarnSchedulerInfo schedulerInfo = null;
public YarnSchedulerInfo getSchedulerInfo() { return schedulerInfo; }
public void collectQueueNames(List<String> queueNames) {
if(schedulerInfo != null) {
schedulerInfo.collectQueueNames(queueNames,null);
}
}
}
@JsonAutoDetect(getterVisibility=Visibility.NONE, setterVisibility=Visibility.NONE, fieldVisibility=Visibility.ANY)
@JsonSerialize(include=JsonSerialize.Inclusion.NON_NULL )
@JsonIgnoreProperties(ignoreUnknown=true)
@XmlRootElement
@XmlAccessorType(XmlAccessType.FIELD)
public static class YarnSchedulerInfo implements java.io.Serializable {
private static final long serialVersionUID = 1L;
private String type = null;
public String getType() {return type;}
private YarnRootQueues rootQueue = null;
public YarnRootQueues getRootQueue() { return rootQueue; }
private String queueName = null;
private YarnQueues queues = null;
public String getQueueName() { return queueName; }
public YarnQueues getQueues() { return queues; }
public void collectQueueNames(List<String> queueNames,String parentQueueName) {
if(type != null && type.equals("fairScheduler")){
if(rootQueue != null) {
rootQueue.collectQueueNames(queueNames);
}
}
else if(type != null && type.equals("capacityScheduler") || type.equals("capacitySchedulerLeafQueueInfo")){
if(queueName != null) {
String queueFqdn = parentQueueName == null ? queueName : parentQueueName + "." + queueName;
queueNames.add(queueFqdn);
if(queues != null) {
queues.collectQueueNames(queueNames, queueFqdn);
}
}
}
else{
try {
throw new Exception("no Schduler");
} catch (Exception e) {
e.printStackTrace();
}
}
}
}
@JsonAutoDetect(getterVisibility=Visibility.NONE, setterVisibility=Visibility.NONE, fieldVisibility=Visibility.ANY)
@JsonSerialize(include=JsonSerialize.Inclusion.NON_NULL )
@JsonIgnoreProperties(ignoreUnknown=true)
@XmlRootElement
@XmlAccessorType(XmlAccessType.FIELD)
public static class YarnRootQueues implements java.io.Serializable {
private static final long serialVersionUID = 1L;
private String queueName = null;
private List<YarnRootQueues> childQueues = null;
public String getQueueName() { return queueName; }
public List<YarnRootQueues> getChildQueues() { return childQueues; }
public void collectQueueNames(List<String> queueNames) {
if(queueName != null){
queueNames.add(queueName);
if(childQueues != null){
for(int i = 0;i < childQueues.size();i++)
queueNames.add(childQueues.get(i).queueName);
}
}
}
}
@JsonAutoDetect(getterVisibility=Visibility.NONE, setterVisibility=Visibility.NONE, fieldVisibility=Visibility.ANY)
@JsonSerialize(include=JsonSerialize.Inclusion.NON_NULL )
@JsonIgnoreProperties(ignoreUnknown=true)
@XmlRootElement
@XmlAccessorType(XmlAccessType.FIELD)
public static class YarnQueues implements java.io.Serializable {
private static final long serialVersionUID = 1L;
private List<YarnSchedulerInfo> queue = null;
public List<YarnSchedulerInfo> getQueue() { return queue; }
public void collectQueueNames(List<String> queueNames, String parentQueueName) {
if(queue != null) {
for(YarnSchedulerInfo schedulerInfo : queue) {
schedulerInfo.collectQueueNames(queueNames, parentQueueName);
}
}
}
}
}