[
https://issues.apache.org/jira/browse/FLINK-15097?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16991085#comment-16991085
]
Paul Lin commented on FLINK-15097:
----------------------------------
+1. I've reported the same issue a while ago,
https://issues.apache.org/jira/browse/FLINK-11135. FYI.
> flink can not use user specified hdfs conf when submitting app in client node
> -----------------------------------------------------------------------------
>
> Key: FLINK-15097
> URL: https://issues.apache.org/jira/browse/FLINK-15097
> Project: Flink
> Issue Type: Bug
> Components: Client / Job Submission
> Affects Versions: 1.9.1
> Reporter: qian wang
> Priority: Major
> Attachments: 0001-adjust-read-hdfs-conf-order.patch
>
>
> now if cluster node had set env HADOOP_CONF_DIR，flink would force use the
> hdfs-site.xml in the corresponding dir, then user who submitted app in the
> client node couldn't use custom specified hdfs-site.xml/hdfs-default through
> setting fs.hdfs.hdfssite or fs.hdfs.hdfsdefault so as to set custom blocksize
> or replication num. For example Using yarnship to upload my hdfs conf dir and
> set fs.hdfs.hdfssite direct to \{conf dir}/hdfs-site.xml is useless
> Deep in code it is due to the order of choosing conf in HadoopUtils.java，the
> conf in HADOOP_CONF_DIR will override user's uploaded conf, i think the way
> is not sensible, so i reverse the order which flink read hdfs conf in order
> to let user custom conf uploaded override HADOOP_CONF_DIR
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
