[
https://issues.apache.org/jira/browse/HDFS-13469?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16459718#comment-16459718
]
Daryn Sharp commented on HDFS-13469:
------------------------------------
-1 on punting. We can't close a serious oversight/regression in the router by
essentially throwing {{TooHardToSupportException}}. It won't be any easier to
fix in the future. That's why I brought up the issue.
Note that's not just the inode path that need to be supported, it's also the
methods with explicit inode-id support. The inode id support was added to make
streams be like file descriptors. Consider the case of a long-lived daemon.
It's writing a file. The file is renamed. Without inode id support, the client
cannot close the file so it lingers in the under construction state. As long
as the client has another file open for writing, or opens another within 1min,
the lease renewal will keep the file indefinitely under construction. We had
this exact problem with various services and it degraded NN performance.
I've long wanted to use inode ids for: directory listings and traversals, temp
files registered for deletion, fetching additional block locations, etc to
improve the correctness of hdfs. The router will break if we use this
legitimate functionality that has been present for many years. That's not
acceptable.
Note that the SPS feature is going to depend on inode paths for directory
traversal, so it'll be incompatible with the router if a solution isn't found.
> RBF: Support InodeID in the Router
> ----------------------------------
>
> Key: HDFS-13469
> URL: https://issues.apache.org/jira/browse/HDFS-13469
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Íñigo Goiri
> Priority: Major
>
> The Namenode supports identifying files through inode identifiers.
> Currently the Router does not handle this properly, we need to add this
> functionality.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
