[ https://issues.apache.org/jira/browse/ARROW-9226?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17303037#comment-17303037 ]
wondertx commented on ARROW-9226: --------------------------------- If HA support is not supported by `fs.HadoopFileSystem`, `pyarrow.hdfs.connect` cannot be simply replaced ` > [Python] pyarrow.fs.HadoopFileSystem - retrieve options from core-site.xml or > hdfs-site.xml if available > -------------------------------------------------------------------------------------------------------- > > Key: ARROW-9226 > URL: https://issues.apache.org/jira/browse/ARROW-9226 > Project: Apache Arrow > Issue Type: Improvement > Components: C++, Python > Affects Versions: 0.17.1 > Reporter: Bruno Quinart > Priority: Minor > Labels: hdfs > Fix For: 4.0.0 > > > 'Legacy' pyarrow.hdfs.connect was somehow able to get the namenode info from > the hadoop configuration files. > The new pyarrow.fs.HadoopFileSystem requires the host to be specified. > Inferring this info from "the environment" makes it easier to deploy > pipelines. > But more important, for HA namenodes it is almost impossible to know for sure > what to specify. If a rolling restart is ongoing, the namenode is changing. > There is no guarantee on which will be active in a HA setup. > I tried connecting to the standby namenode. The connection gets established, > but when writing a file an error is raised that standby namenodes are not > allowed to write to. > -- This message was sent by Atlassian Jira (v8.3.4#803005)