Hi devs,
I play with your amazing Spark here in Prague for some time. I have
stumbled on a thing which I like to ask about. I create assembly jars from
source and then use it to run simple jobs on our 2.3.0-cdh5.1.3 cluster
using yarn. Example of my usage [1]. Formerly I had started to use sbt for
creating assemblies like this [2] which runs just fine. Then reading those
maven-prefered stories here on dev list I found make-distribution.sh script
in root of codebase and wanted to give it a try. I used it to create
assembly by both [3] and [4].
But I am not able to use assemblies created by make-distribution because
it refuses to be submited to cluster. Here is what happens:
- run [3] or [4]
- recompile app agains new assembly
- submit job using new assembly by [1] like command
- submit fails with important parts of stack trace being [5]
My guess is that it is due to improper version of protobuf included in
assembly jar. My questions are:
- Can you confirm this hypothesis?
- What is the difference between sbt and mvn way of creating assembly? I
mean sbt works and mvn not...
- What additional option I need to pass to make-distribution to make it
work?
Any help/explanation here would be appreciated
Jakub
----------------------
[1] ./bin/spark-submit --num-executors 200 --master yarn-cluster --conf
spark.yarn.jar=assembly/target/scala-2.10/spark-assembly-1.2.1-SNAPSHOT-
hadoop2.3.0-cdh5.1.3.jar --class org.apache.spark.mllib.
CreateGuidDomainDictionary root-0.1.jar ${args}
[2] ./sbt/sbt -Dhadoop.version=2.3.0-cdh5.1.3 -Pyarn -Phive assembly/
assembly
[3] ./make-distribution.sh -Dhadoop.version=2.3.0-cdh5.1.3 -Pyarn -Phive -
DskipTests
[4] ./make-distribution.sh -Dyarn.version=2.3.0 -Dhadoop.version=2.3.0-cdh
5.1.3 -Pyarn -Phive -DskipTests
[5]Exception in thread "main" org.apache.hadoop.yarn.exceptions.
YarnRuntimeException: java.lang.reflect.InvocationTargetException
at org.apache.hadoop.yarn.factories.impl.pb.RpcClientFactoryPBImpl.
getClient(RpcClientFactoryPBImpl.java:79)
at org.apache.hadoop.yarn.ipc.HadoopYarnProtoRPC.getProxy
(HadoopYarnProtoRPC.java:48)
at org.apache.hadoop.yarn.client.RMProxy$1.run(RMProxy.java:134)
...
Caused by: java.lang.reflect.InvocationTargetException
at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native
Method)
at sun.reflect.NativeConstructorAccessorImpl.newInstance
(NativeConstructorAccessorImpl.java:39)
at sun.reflect.DelegatingConstructorAccessorImpl.newInstance
(DelegatingConstructorAccessorImpl.java:27)
at java.lang.reflect.Constructor.newInstance(Constructor.java:513)
at org.apache.hadoop.yarn.factories.impl.pb.RpcClientFactoryPBImpl.
getClient(RpcClientFactoryPBImpl.java:76)
... 27 more
Caused by: java.lang.VerifyError: class org.apache.hadoop.yarn.proto.
YarnServiceProtos$SubmitApplicationRequestProto overrides final method
getUnknownFields.()Lcom/google/protobuf/UnknownFieldSet;
at java.lang.ClassLoader.defineClass1(Native Method)
at java.lang.ClassLoader.defineClassCond(ClassLoader.java:631)
