[
https://issues.apache.org/jira/browse/SPARK-15949?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15330060#comment-15330060
]
Eliano Marques commented on SPARK-15949:
----------------------------------------
Sean, thanks for your reply, I will read carefully your link. Please note that
before I started this issue I investigated a few of the reasons of such
behaviour and it's not the case I'm running with a wrong user.
Both APIs are trying to start the job into an incorrect HDFS location and they
fail accordingly. I can provide you with the details of the error but as the
behaviour in the spark-shell being different i though this was something you
guys wanted to check. As it stands the hive context is not really working.
> Spark 2.0.
> ----------
>
> Key: SPARK-15949
> URL: https://issues.apache.org/jira/browse/SPARK-15949
> Project: Spark
> Issue Type: Bug
> Components: PySpark, SparkR
> Affects Versions: 2.0.0
> Environment: Hadoop 2.6 with cloudera
> Reporter: Eliano Marques
>
> When using pyspark or sparkR to connect to Hive there is a permissions error
> back to hdfs. In the beginning it sounded like a permissions problem but we
> realise this might have to do with the APIs itself since with scala this
> issue doesn't show up.
> Pyspark:
> spark and sc is available and you can see the new SparkSession.
> when you do:
> sqlContext("show databases").show()
> An error shows due to permissions in "hdfs://nameservice1/home/username/etc"/
> The same behaviour in SparkR with:
> databases = sql(sqlContext,"show databases"))
> In scala (spark-shell) the operation sqlContext("show databases") correctly
> goes to "hdfs://nameservice1/user/username" which is where all hfds files for
> a specific user should be.
> Let me know if you need more details.
> Thanks
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
