One more question on yarn-mode.

After setting Spark Yarn Jar path and Spark Home path in Spark interpreter
(via UI) I was expecting Zeppelin to use my own version of Spark. I don't
think that is happening. From zeppelin, if I do sc.version it gives 1.2.1
(which is included for CDH 5.3). When I run spark-shell from my custom
Spark lib, version shows as 1.2.0.

I have set Spark Yarn Jar, Spark Home (from Interpreter UI) and Hadoop conf
dir (in zeppelin-env.sh). Do I need to do anything else for zeppelin to use
my spark jar?

Also, my hive context does not recognize the databases/tables on the
cluster. Do I need to point or copy hive-site.xml anywhere in zeppelin's
conf?

Thanks,
Charmee



On Tue, Apr 14, 2015 at 9:32 PM Jongyoul Lee <[email protected]> wrote:

> Good!!
>
> On Tue, Apr 14, 2015 at 11:14 PM, Charmee Patel <[email protected]>
> wrote:
>
>> Hi,
>>
>> Thanks.
>>
>> I had pulled the code from https://github.com/NFLabs/zeppelin repository
>> when it had not moved to apache/incubator-zeppelin yet and I kept pulling
>> from same repo. Synced up my version with latest on
>> apache/incubator-zeppelin and it is working now.
>>
>> -Charmee
>>
>> On Mon, Apr 13, 2015 at 9:12 PM Jongyoul Lee <[email protected]> wrote:
>>
>>> Hi,
>>>
>>> I don't know what version you use exactly because com.nflabs.zeppelin
>>> moves to org.apache.zeppelin. You don't need to set SPARK_YARN_JAR at the
>>> latest master. Could you please check this out? And I've test yarn-client
>>> mode on 2.5.0-cdh5.3.0 with spark 1.3 but didn't do it with spark 1.2. My
>>> build script is
>>>
>>> mvn clean package -Pspark-1.3 -Dhadoop.version=2.5.0-cdh5.3.0
>>> -Phadoop-2.4 -DskipTests -Pyarn -Pbuild-distr
>>>
>>> Regards,
>>> Jongyoul Lee
>>>
>>> On Mon, Apr 13, 2015 at 11:40 PM, Charmee Patel <[email protected]>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> I am trying to get Zeppelin work in yarn-client and yarn-submit
>>>> mode. There are some conflicting notes in the email distros about how to
>>>> get Zeppelin to work on yarn-client mode. I have tried a few different
>>>> things but none have worked for me so far.
>>>>
>>>> I have CDH 5.3 . Here is what I have so far
>>>>
>>>>    1. Built Zeppelin on local (OSX) using
>>>>    1. mvn clean package -Pspark-1.2 -Dhadoop.version=2.5.0-cdh5.3.0
>>>>       -Phadoop-2.4 -DskipTests -Pyarn
>>>>    2. Generated a distribution package and deployed to edge node of my
>>>>    cluster using
>>>>    1. mvn clean package -P build-distr -DskipTests
>>>>    3. At this point local mode works fine
>>>>    4. To get Yarn mode working
>>>>       1. I set Hadoop conf dir in zeppelin-env.sh and set Yarn
>>>>       Master=yarn-client
>>>>       I get an exception in my logs
>>>>
>>>>       ERROR ({pool-2-thread-5} Job.java[run]:165) - Job failed
>>>>       com.nflabs.zeppelin.interpreter.InterpreterException:
>>>>       org.apache.thrift.TApplicationException: Internal error processing 
>>>> open
>>>>
>>>>       2. Set Spark_Yarn_Jar but got same exception as above
>>>>       3. Copied my spark assembly jar in interpreter/spark directory
>>>>       but that did not work either
>>>>       4. Also set Spark Yarn Jar and Spark Home in interpreter UI but
>>>>       that did not work
>>>>       5. I took Spark Yarn Jar out and let Zeppelin use Spark that
>>>>       comes bundled with it
>>>>          1. This seemed to work initially, but as soon as I call Spark
>>>>          Action (collect/count etc) I get this exception
>>>>
>>>> java.lang.RuntimeException: Error in configuring object at
>>>> org.apache.hadoop.util.ReflectionUtils.setJobConf(ReflectionUtils.java:93)
>>>> at org.apache.hadoop.util.ReflectionUtils.setConf(ReflectionUtils.java:64)
>>>> at
>>>> org.apache.hadoop.util.ReflectionUtils.newInstance(ReflectionUtils.java:117)
>>>> at org.apache.spark.rdd.HadoopRDD.getInputFormat(HadoopRDD.scala:182)
>>>>
>>>> Any pointers on I might have missed in configuring zeppelin to work
>>>> with yarn-client mode?
>>>>
>>>> Thanks,
>>>> Charmee
>>>>
>>>
>>>
>>>
>>> --
>>> 이종열, Jongyoul Lee, 李宗烈
>>> http://madeng.net
>>>
>>
>
>
> --
> 이종열, Jongyoul Lee, 李宗烈
> http://madeng.net
>

Reply via email to