Hi all,
I think there is an issue in cooperation of HftpFileSystem (hftp://) and
HDFS High Availability. The read might fail in the following scenario:
* a cluster is configured in HA mode, with the following configuration:
<property>
<name>dfs.nameservices</name>
<value>master</value>
</property>
<property>
<name>dfs.ha.namenodes.master</name>
<value>master1,master2</value>
</property>
...
* 'master1' is set to standby and 'master2' to active
* the following command gives error
$ hadoop fs -ls hftp://master/
ls: Operation category READ is not supported in state standby
* while the following succeeds:
$ hadoop fs -ls hdfs://master/
Found 98 items
...
I have not checked the code, but I have a suspicion, that the
HftpFileSystem takes always the first host, or it doesn't handle the
thrown exception correctly.
Is this a known issue? Should the error be handled in some kind of
wrapper in client code, or is there some other workaround? Should this
be fixed in the HftpFileSystem (somehow)?
Thanks for opinions,
Jan