CODEWITHSUNDEEP
apache-sparkoozieemroozie-coordinator
Oscar Daniel
Oscar Daniel

Reputation: 21

Spark job in oozie coordinator error - emr: Can not create a Path from an empty string

I have a problem configuring a coordinator with oozie in a yarn cluster, it's an spark job, when I run the workflow by console the job is launched and executed correctly by the yarn, but when i call the same workflow from an coordinator.xml i have this error:

ERROR org.apache.spark.SparkContext  - Error initializing SparkContext.
java.lang.IllegalArgumentException: Can not create a Path from an empty    string
    at org.apache.hadoop.fs.Path.checkPathArg(Path.java:127)
    at org.apache.hadoop.fs.Path.<init>(Path.java:135)
    at org.apache.hadoop.fs.Path.<init>(Path.java:94)
    at org.apache.spark.deploy.yarn.Client.copyFileToRemote(Client.scala:337)

And the job never is launched in the yarn cluster, looks like yarn can't receive the .jar correct path from oozie, any idea?

Here the coordinator.xml and the workflow.xml simplified.

<coordinator-app name="Firebase acquisition process coordinator" frequency="${coord:days(1)}"
start="${startTime}" end="${endTime}" timezone="UTC" xmlns="uri:oozie:coordinator:0.5">
   <controls>
...
   </controls>
   <action>
      <workflow>
         <app-path>hdfs://ip-111-11-11-111.us-west-  2.compute.internal:8020/user/hadoop/emr-spark/</app-path>
      </workflow>
   </action>
</coordinator-app>

<workflow-app name="bbbbbbbbbbbbbbb" xmlns="uri:oozie:workflow:0.5">
    <start to="spark-0324"/>
    <kill name="Kill">
        <message>Action failed, error message[${wf:errorMessage(wf:lastErrorNode())}]</message>
    </kill>
    <action name="spark-0324">
        <spark xmlns="uri:oozie:spark-action:0.2">
            <job-tracker>${jobTracker}</job-tracker>
            <name-node>${nameNode}</name-node>
            <master>yarn</master>
            <mode>client</mode>
              <class>classsxxx.Process</class>
            <jar>hdfs://ip-111-11-11-111.us-west-2.compute.internal:8020/user/hadoop/emr-spark/lib/jarnamex.jar</jar>
            <file>lib#lib</file>
        </spark>
        <ok to="End"/>
        <error to="Kill"/>
    </action>
    <end name="End"/>
</workflow-app>

I mean, when I do this; oozie job -config ~/emr-spark/job.properties -run it works!!, but when I try this; oozie job -run -config ~/emr-coordinator/coordinator.properties It doesn't work.

job properties

oozie.use.system.libpath=true
send_email=False
dryrun=False
nameNode=hdfs://ip-111-11-11-111.us-west-2.compute.internal:8020
jobTracker=ip-111-11-11-111.us-west-2.compute.internal:8032
oozie.wf.application.path=/user/hadoop/emr-spark

coordinator properties

startTime=2017-09-08T19:46Z
endTime=2030-01-01T06:00Z
jobTracker=ip-111-11-11-111.us-west-2.compute.internal:8032
nameNode=hdfs://ip-111-11-11-111.us-west-2.compute.internal:8020
oozie.coord.application.path=hdfs://ip-111-11-11-111.us-west-2.compute.internal:8020/user/hadoop/emr-coordinator
oozie.use.system.libpath=true

Upvotes: 2

Views: 1024

Answers (1)

Oscar Daniel
Oscar Daniel

Reputation: 21

Referring to resource from the HDFS file system it has to be relative only. The full/absolute path is computed on demand.

Then the solution was just replace: hdfs://ip-111-11-11-111.us-west-2.compute.internal:8020/user/hadoop/emr-spark/workflow.xml with: /user/hadoop/emr-spark/workflow.xml and hdfs://ip-111-11-11-111.us-west-2.compute.internal:8020/user/hadoop/emr-spark/lib/xxxx.jar with /user/hadoop/emr-spark/lib/xxxxx.jar

In the workflow.xml, coordinator.xml or properties.

Upvotes: 0

Related Questions