本文以一个完整的程序为例,提交一个计算π的作业到数据湖分析DLA,跟踪它的状态,查询历史上的运行结果。
import com.aliyuncs.DefaultAcsClient;
import com.aliyuncs.IAcsClient;
import com.aliyuncs.exceptions.ClientException;
import com.aliyuncs.openanalytics_open.model.v20180619.*;
import com.aliyuncs.profile.DefaultProfile;
import java.io.IOException;
import java.util.Arrays;
import java.util.List;
/**
* 演示如何使用Java SDK操作数据湖分析的spark作业
*
* @author aliyun
*/
public class Demo {
/**
* 提交一个SQL作业到数据湖分析Serverless Spark
*
* @param virtualClusterName 数据湖分析虚拟集群名称
* @param sql sql内容
* @return Spark JobId, 提交作业成功, 返回作业的ID, 用于后续的状态跟踪
* @throws ClientException 提交作业可能因为网络原因等抛出错误
*/
public static String submitSparkSQL(IAcsClient client,
String virtualClusterName,
String sql) throws ClientException {
// 初始化Request, 填入集群名称和作业内容
SubmitSparkSQLRequest request = new SubmitSparkSQLRequest();
request.setVcName(virtualClusterName);
request.setSql(sql);
// 提交作业, 返回Spark作业的JobId
SubmitSparkSQLResponse response = client.getAcsResponse(request);
return response.getJobId();
}
/**
* 提交一个作业到数据湖分析Serverless Spark
*
* @param virtualClusterName 数据湖分析虚拟集群名称
* @param jobConfig 提交Spark作业的描述文件,需要是JSON格式
* @return Spark JobId, 提交作业成功, 返回作业的ID, 用于后续的状态跟踪
* @throws ClientException 提交作业可能因为网络原因等抛出错误
*/
public static String submitSparkJob(IAcsClient client,
String virtualClusterName,
String jobConfig) throws ClientException {
// 初始化Request, 填入集群名称和作业内容
SubmitSparkJobRequest request = new SubmitSparkJobRequest();
request.setVcName(virtualClusterName);
request.setConfigJson(jobConfig);
// 提交作业, 返回Spark作业的JobId
SubmitSparkJobResponse response = client.getAcsResponse(request);
return response.getJobId();
}
/**
* 返回一个Spark Job当前的状态
*
* @param sparkJobId 用户Spark作业的ID
* @return 返回Spark作业的状态, 类型为String
* @throws ClientException 提交作业可能因为网络原因等抛出错
*/
public static String getSparkJobStatus(IAcsClient client,
String virtualClusterName,
String sparkJobId) throws ClientException {
// 初始化Request, 填入spark job id
GetJobStatusRequest request = new GetJobStatusRequest();
request.setJobId(sparkJobId);
request.setVcName(virtualClusterName);
// 提交作业, 返回Spark作业的状态码
GetJobStatusResponse response = client.getAcsResponse(request);
return response.getStatus();
}
/**
* 停止一个Spark Job
*
* @param sparkJobId 用户Spark作业的ID
* @param virtualClusterName 数据湖分析虚拟集群名称
* @return 无返回值
* @throws ClientException 提交作业可能因为网络原因等抛出错
*/
public static void killSparkJob(IAcsClient client,
String virtualClusterName,
String sparkJobId) throws ClientException {
// 初始化Request, 填入spark job id
KillSparkJobRequest request = new KillSparkJobRequest();
request.setVcName(virtualClusterName);
request.setJobId(sparkJobId);
// 提交作业, 返回Spark作业的状态码
KillSparkJobResponse response = client.getAcsResponse(request);
}
/**
* 返回一个Spark Job的日志
*
* @param client 客户端
* @param virtualClusterName 数据湖分析虚拟集群名称
* @param sparkJobId 用户Spark作业的ID
* @return 返回Spark作业的状态, 类型为String
* @throws ClientException 提交作业可能因为网络原因等抛出错
*/
public static String getSparkJobLog(IAcsClient client,
String virtualClusterName,
String sparkJobId) throws ClientException {
// 初始化Request, 填入spark job id
GetJobLogRequest request = new GetJobLogRequest();
request.setJobId(sparkJobId);
request.setVcName(virtualClusterName);
// 提交作业, 返回Spark作业的日志
GetJobLogResponse response = client.getAcsResponse(request);
return response.getData();
}
/**
* 查询某个虚拟集群上提交的Spark作业, 通过翻页可以遍历所有的历史作业信息
*
* @param client 客户端
* @param pageNumber 查询的页码, 从1开始
* @param pageSize 每页返回数量
* @throws ClientException 提交作业可能因为网络原因等抛出错
*/
public static void listSparkJob(IAcsClient client,
String virtualClusterName,
int pageNumber,
int pageSize) throws ClientException {
// 初始化Request, 填入spark job id
ListSparkJobRequest request = new ListSparkJobRequest();
request.setVcName(virtualClusterName);
request.setPageNumber(pageNumber); // pageNumber 从1开始
request.setPageSize(pageSize);
// 提交作业, 返回Spark作业的状态码
ListSparkJobResponse response = client.getAcsResponse(request);
// 获取任务列表
List<ListSparkJobResponse.DataResult.Data> sparkJobList = response.getDataResult().getJobList();
for (ListSparkJobResponse.DataResult.Data job : sparkJobList) {
System.out.println(String.format("JobName: %s, JobUI: %s, JobStatus: %s, JobConfig: %s",
job.getJobName(),
job.getStatus(),
job.getSparkUI(),
job.getDetail()));
}
}
public static void main(String[] args) throws IOException, ClientException, InterruptedException {
// 提交任务必须的参数
String region = "cn-hangzhou";
// 获取一个合法的AK SK
String accessKeyId = System.getEnv("RAM_AK");
String accessKeySecret = "System.getEnv("RAM_SK");
String virtualClusterName = "MyCluster";
// 需要是一个合法的JSON格式的字符串
String jobConfig=
"{\n" +
" \"name\": \"SparkPi\",\n" +
" \"file\": \"local:///tmp/spark-examples.jar\",\n" +
" \"className\": \"org.apache.spark.examples.SparkPi\",\n" +
" \"args\": [\n" +
" \"100\"\n" +
" ],\n" +
" \"conf\": {\n" +
" \"spark.driver.resourceSpec\": \"medium\",\n" +
" \"spark.executor.instances\": 5,\n" +
" \"spark.executor.resourceSpec\": \"medium\"\n" +
" }\n" +
"}";
String sql = "-- here is the spark conf\n"
+ "set spark.driver.resourceSpec=medium;\n"
+ "set spark.executor.instances=5;\n"
+ "set spark.executor.resourceSpec=medium;\n"
+ "set spark.app.name=sparksqltest;\n"
+ "set spark.sql.hive.metastore.version=dla;\n"
+ "-- here is your sql statement\n"
+ "-- add your jar\n"
+ "-- add jar oss://path/to/your/jar\n"
+ "show databases;";
// 初始化阿里云平台开发Client
DefaultProfile profile = DefaultProfile.getProfile(region, accessKeyId, accessKeySecret);
IAcsClient client = new DefaultAcsClient(profile);
// 提交任务
// 您也可以选择提交SQL作业
//String sparkJobId = submitSparkSQL(client, virtualClusterName, sql);
String sparkJobId = submitSparkJob(client, virtualClusterName, jobConfig);
// 轮询任务状态, 超时未完成则杀死任务
long startTime = System.currentTimeMillis();
List<String> finalStatusList = Arrays.asList("error", "success", "dead", "killed");
while (true) {
String status = getSparkJobStatus(client, virtualClusterName, sparkJobId);
if (finalStatusList.contains(status)) {
System.out.println("Job went to final status");
break;
} else if ((System.currentTimeMillis() - startTime) > 100000) {
// 如果超时则杀死job
System.out.println("Kill expire time job");
killSparkJob(client, virtualClusterName, sparkJobId);
break;
}
// 打印状态, 等待5秒, 进入下一轮查询
System.out.println(String.format("Job %s status is %s", sparkJobId, status));
Thread.sleep(5000);
}
// 打印作业的日志
String logDetail = getSparkJobLog(client, virtualClusterName, sparkJobId);
System.out.println(logDetail);
// 打印最近10条作业的明细
listSparkJob(client, virtualClusterName, 1, 10);
}
}