Hadoop源码分析28 JobTracker 处理JobClient请求

提交命令行：hadoop jar/opt/hadoop-1.0.0/hadoop-examples-1.0.0.jar wordcount/user/admin/in/yellow.txt /user/admin/out/3

 

RPC请求：getProtocolVersion(org.apache.hadoop.mapred.JobSubmissionProtocol,28) from 10.1.1.101:37721

处理：返回JobSubmitProtocol的Version

返回：28

 

RPC请求：getStagingAreaDir()from 10.1.1.101:37722

处理：查看conf.get("mapreduce.jobtracker.staging.root.dir")

返回：hdfs://server1:9000/tmp/hadoop-admin/mapred/staging/admin/.staging

 

RPC请求：getNewJobId()from 10.1.1.101:37725

处理：new JobID(getTrackerIdentifier(),nextJobId++)

返回：job_201404152128_0001

 

RPC请求：getQueueAdmins(default)from 10.1.1.101:37730

处理：aclsEnabled=false, 故allAllowed=true     

返回：All usersare allowed

 

RPC请求：submitJob(job_201404152128_0001,hdfs://server1:9000/tmp/hadoop-admin/mapred/staging/admin/.staging/job_201404152128_0001,org.apache.hadoop.security.Credentials@6e28575) from10.1.1.101:37734

处理：

生成JobInfo对象：{id=job_201404152128_0001，jobSubmitDir=hdfs://server1:9000/tmp/hadoop-admin/mapred/staging/admin/.staging/job_201404152128_0001，user=admin}

生成JobInProgress对象：其中包括一个JobStatus对象{jobid=job_201404152128_0001;mapProgress=0; reduceProgress=0; cleanupProgress=0; setupProgress=0; runState=4;  startTime=1397612941994; user=admin; priority=NOMAL; schedulingInfo="NA"; failureInfo = "NA"; } 

将文件job.xml从HDFS复制到本地：fs.copyToLocalFile(jobFilePath,localJobFile);

（jobFilePath=hdfs://server1:9000/tmp/hadoop-admin/mapred/staging/admin/.staging/job_201404152128_0001/job.xml，/tmp/hadoop-admin/mapred/local/jobTracker/job_201404152128_0001.xml）

将该文件加载到配置：conf=newJobConf(localJobFile);

然后读取job.xml中的配置

生成JobProfile对象：{jobFile="hdfs://server1:9000/tmp/hadoop-admin/mapred/staging/admin/.staging/job_201404152128_0001/job.xml"，

jobid=job_201404152128_0001，name="wordcount"，queueName="default"，url="http://server1:50030/jobdetails.jsp?jobid=job_201404152128_0001"(id=720)   

user="admin"}

totalSubmissions++，

将JobInProgress添加到jobs，其内容为

{job_201404152128_0001=org.apache.hadoop.mapred.JobInProgress@7e9b4b1f} 

将JobInProgress添加到JobQueueJobInProgressListener、EagerTaskInitializationListener

 

EagerTaskInitializationListener线程从jobInitQueue取出JobInProgress，初始化JobInProgress：

threadPool.execute(new InitJob(job));

实际调用JobTracker.initJob(JobInProgressjob)

从hdfs://server1:9000/tmp/hadoop-admin/mapred/staging/admin/.staging/job_201404152128_0001/job.splitmetainfo读取TaskSplitMetaInfo，生成相应的TaskInProgress(maps、reduces、cleanup、setup)

返回：JobInProgress的成员JobStatus

 

RPC请求：getJobProfile(job_201404152128_0001)from 10.1.1.101:37744

处理：

返回：JobInProgress的成员JobProfile

 

RPC请求：getJobStatus(job_201404152128_0001)from 10.1.1.101:37744

处理：

返回：JobInProgress的成员JobStatus

 

RPC请求：getTaskCompletionEvents(job_201404152128_0001,0, 10) from 10.1.1.101:37744

处理：

返回：JobInProgress的成员taskCompletionEvents

..................

此后JobClient会不断请求getJobStatus、getTaskCompletionEvents