[
https://issues.apache.org/jira/browse/HDFS-11437?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16078556#comment-16078556
]
James Clampffer commented on HDFS-11437:
----------------------------------------
Proposing a short term workaround in HDFS-12103. Once this is fixed that could
be removed (if workaround ends up being committed).
> libhdfs++: Handler for FileSystem async connect can be invoked before
> successful communication with active NN
> -------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-11437
> URL: https://issues.apache.org/jira/browse/HDFS-11437
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: hdfs-client
> Reporter: James Clampffer
> Assignee: James Clampffer
>
> The handler provided to FileSystem::Connect can be invoked as soon as the FS
> makes a connection to the standby NN rather than waiting until it connects to
> the active NN. This allows RPC requests to be enqueued before a real
> connection is made and if the active NN isn't reachable for some reason the
> only way to cancel is to delete the FS from another thread which kills all
> pending requests.
> The underlying issue is that currently the only thing that must happen for
> the connect handler to be invoked is a successful handshake with a NN.
> Connecting to the standby NN and receiving a StandbyException satisfies this
> requirement but it should wait until it is able to get a handshake from the
> active NN.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
