[jira] [Commented] (MESOS-3527) HDFS HA fails outside of docker context

["Timothy Chen (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MESOS-3527?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15032505#comment-15032505 ] 

Timothy Chen commented on MESOS-3527:
-------------------------------------

Hi Alan, the fetcher in Mesos has always been running outside of the docker container, since the fetcher itself depends on libmesos.

Since hdfs needs core-site.xml/hdfs config information, is those configuration available in the container you're running or on the slave? 



> HDFS HA fails outside of docker context
> ---------------------------------------
>
>                 Key: MESOS-3527
>                 URL: https://issues.apache.org/jira/browse/MESOS-3527
>             Project: Mesos
>          Issue Type: Bug
>            Reporter: Alan Braithwaite
>              Labels: fetcher, hdfs, mesosphere, spark
>
> I'm using Spark with the Mesos driver.
> When I pass in a `hdfs://<namespace>/path` url in for the spark application, the fetcher attempts to download the jar files outside the spark context (the docker container in this case).  The problem is that the core-site.xml and hdfs-site.xml configs exist inside the container.  The host machine does not have the necessary hdfs configuration to connect to the HA cluster.
> Currently, I'm not sure what the alternative ways of accessing a HA hadoop cluster besides through the hadoop client.
> {code}
> I0926 06:34:19.346851 18851 fetcher.cpp:214] Fetching URI 'hdfs://hdfsha/tmp/spark-job.jar'
> I0926 06:34:19.622860 18851 fetcher.cpp:99] Fetching URI 'hdfs://hdfsha/tmp/spark-job.jar' using Hadoop Client
> I0926 06:34:19.622936 18851 fetcher.cpp:109] Downloading resource from 'hdfs://hdfsha/tmp/spark-job.jar' to '/state/var/lib/mesos/slaves/20150602-065056-269165578-5050-17724-S12/frameworks/20150914-102037-285942794-5050-31214-0029/executors/driver-20150926063418-0002/runs/9953ae1b-9387-489f-8645-5472d9c5eacf/spark-job.jar'
> E0926 06:34:20.814858 18851 fetcher.cpp:113] HDFS copyToLocal failed: /usr/local/hadoop/bin/hadoop fs -copyToLocal 'hdfs://hdfsha/tmp/spark-job.jar' '/state/var/lib/mesos/slaves/20150602-065056-269165578-5050-17724-S12/frameworks/20150914-102037-285942794-5050-31214-0029/executors/driver-20150926063418-0002/runs/9953ae1b-9387-489f-8645-5472d9c5eacf/spark-job.jar'
> -copyToLocal: java.net.UnknownHostException: hdfsha
> Usage: hadoop fs [generic options] -copyToLocal [-p] [-ignoreCrc] [-crc] <src> ... <localdst>
> Failed to fetch: hdfs://hdfsha/tmp/spark-job.jar
> {code}
> The code in question:
> https://github.com/apache/mesos/blob/fbb12a52969710fe69c309c83db0a5441dbea886/src/launcher/fetcher.cpp#L92-L114



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)