【原創】大叔經驗分享(6)Oozie如何查看提交到Yarn上的任務日志


通過oozie job id可以查看流程詳細信息,命令如下:

oozie job -info 0012077-180830142722522-oozie-hado-W

 

流程詳細信息如下:

Job ID : 0012077-180830142722522-oozie-hado-W

------------------------------------------------------------------------------------------------------------------------------------

Workflow Name : test_wf

App Path      : hdfs://hdfs_name/oozie/test_wf.xml

Status        : KILLED

Run           : 0

User          : hadoop

Group         : -

Created       : 2018-09-25 02:51 GMT

Started       : 2018-09-25 02:51 GMT

Last Modified : 2018-09-25 02:53 GMT

Ended         : 2018-09-25 02:53 GMT

CoordAction ID: -

 

Actions

------------------------------------------------------------------------------------------------------------------------------------

ID                                                                            Status    Ext ID                 Ext Status Err Code 

------------------------------------------------------------------------------------------------------------------------------------

0012077-180830142722522-oozie-hado-W@:start:                                  OK        -                      OK         -        

------------------------------------------------------------------------------------------------------------------------------------

0012077-180830142722522-oozie-hado-W@test_spark_task                  ERROR     application_1537326594090_5663FAILED/KILLEDJA018    

------------------------------------------------------------------------------------------------------------------------------------

0012077-180830142722522-oozie-hado-W@Kill                                     OK        -                      OK         E0729    

------------------------------------------------------------------------------------------------------------------------------------

 

失敗的任務定義如下

<action name="test_spark_task"> 

        <spark xmlns="uri:oozie:spark-action:0.1"> 

            <job-tracker>${job_tracker}</job-tracker> 

            <name-node>${name_node}</name-node> 

            <master>${jobmaster}</master> 

            <mode>${jobmode}</mode> 

            <name>${jobname}</name> 

            <class>${jarclass}</class> 

            <jar>${jarpath}</jar> 

            <spark-opts>--executor-memory 4g --executor-cores 2 --num-executors 4 --driver-memory 4g</spark-opts> 

        </spark>

 

在yarn上可以看到application_1537326594090_5663對應的application如下

application_1537326594090_5663       hadoop oozie:launcher:T=spark:W=test_wf:A=test_spark_task:ID=0012077-180830142722522-oozie-hado-W         Oozie Launcher

 

查看application_1537326594090_5663日志發現

2018-09-25 10:52:05,237 [main] INFO  org.apache.hadoop.yarn.client.api.impl.YarnClientImpl  - Submitted application application_1537326594090_5664

 

yarn上application_1537326594090_5664對應的application如下

application_1537326594090_5664       hadoop    TestSparkTask SPARK

 

即application_1537326594090_5664才是Action對應的spark任務,為什么中間會多一步,類結構和核心代碼詳見 https://www.cnblogs.com/barneywill/p/9895225.html

簡要來說,Oozie執行Action時,即ActionExecutor(最主要的子類是JavaActionExecutor,hive、spark等action都是這個類的子類),JavaActionExecutor首先會提交一個LauncherMapper(map任務)到yarn,其中會執行LauncherMain(具體的action是其子類,比如JavaMain、SparkMain等),spark任務會執行SparkMain,在SparkMain中會調用org.apache.spark.deploy.SparkSubmit來提交任務

 

如果提交的是spark任務,那么按照上邊的方法就可以跟蹤到實際任務的applicationId;
如果你提交的hive2任務,實際是用beeline啟動,從hive2開始,beeline命令的日志已經簡化,不像hive命令可以看到詳細的applicationId和進度,這時有兩種方法:

1)修改hive代碼,使得beeline命令和hive命令一樣有詳細日志輸出

詳見:https://www.cnblogs.com/barneywill/p/10185949.html

2)根據application tag手工查找任務

oozie在使用beeline提交任務時,會添加一個mapreduce.job.tags參數,比如

--hiveconf
mapreduce.job.tags=oozie-9f896ad3d40c261235dc6858cadb885c

但是這個tag從yarn application命令中查不到,只能手工逐個查找(實際啟動的任務會在當前LuancherMapper的applicationId上遞增),

然后就可以看到實際啟動的applicationId了

 

另外還可以從job history server上看到application的詳細信息,比如configuration、task等

查看hive任務執行的完整sql詳見:https://www.cnblogs.com/barneywill/p/10083731.html

 


免責聲明!

本站轉載的文章為個人學習借鑒使用,本站對版權不負任何法律責任。如果侵犯了您的隱私權益,請聯系本站郵箱yoyou2525@163.com刪除。



 
粵ICP備18138465號   © 2018-2025 CODEPRJ.COM