[
https://issues.apache.org/jira/browse/FLINK-7951?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16234177#comment-16234177
]
ASF GitHub Bot commented on FLINK-7951:
---------------------------------------
Github user StephanEwen commented on the issue:
https://github.com/apache/flink/pull/4926
@steveloughran if it is okay, maybe we could pick could brain here quickly?
What is the Yarn-ideomatic way to handle configurations and what are the
assumptions?
Specifically:
- Do ApplicationMaster and worker processes launched by Yarn typically
assume that all configurations (`core-site.xml`, `hdfs-site.xml`,
`yarn-site.xml`) are in the classpath, or should they attempt to load
configurations by taking environment variables into account?
- Related to that, should the "client" process that submits the yarn
application attach all configurations that it finds as resources, or not attach
and configuration files, assuming that the processes launched by the node
managers find their own configurations on the yarn cluster nodes?
> YarnApplicationMaster does not load HDFSConfiguration
> -----------------------------------------------------
>
> Key: FLINK-7951
> URL: https://issues.apache.org/jira/browse/FLINK-7951
> Project: Flink
> Issue Type: Bug
> Components: YARN
> Affects Versions: 1.4.0
> Reporter: Till Rohrmann
> Assignee: Till Rohrmann
> Priority: Critical
> Fix For: 1.4.0
>
>
> When instantiating the {{YarnConfiguration}} we do not load the corresponding
> {{HDFSConfiguration}}. This causes that we do not read the {{hdfs-site.xml}}.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
