[
https://issues.apache.org/jira/browse/HDFS-6765?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14078277#comment-14078277
]
Daryn Sharp commented on HDFS-6765:
-----------------------------------
My bad, I forgot 2.x was more FD-ish. I actually filed jiras long ago about
the rename/close/orphaned lease problem so I'm glad that's fixed.
I was vague regarding oev, I was referring to backwards-compatibility wrt to
usability. Any tool that depends on oev records containing a path will be
broken. As a human, I'd rather have the path in plain view versus rummaging
through previous edit log segments or ultimately an image.
AddCloseOp already contains an inode field that isn't being set. Isn't the
simplest & compatible solution to set the inode field and use the id with
preference over the path? It won't require another layout change and will
enable rolling back to an earlier NN if there's a nasty bug in a new NN.
> Log INodeID instead of full path in edit log when closing a file
> ----------------------------------------------------------------
>
> Key: HDFS-6765
> URL: https://issues.apache.org/jira/browse/HDFS-6765
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Haohui Mai
>
> When closing a file, currently the edit log records the full path instead of
> the inode id. The performance is sub-optimal, and it prevents further clean
> ups of the code (e.g., HDFS-6757)
> This jira proposes to record inode id instead of the full path in edit logs
> when closing a file.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
