[jira] Commented: (HADOOP-1774) Remove use of INode.parent in Block CRC upgrade

["Konstantin Shvachko (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-1774?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12526324 ] 

Konstantin Shvachko commented on HADOOP-1774:
---------------------------------------------

The code looks good to me. The name INodeWrapper sound to generic.
I'd propose InodeParentPair, or ParentMapEntry, or simply INodeParent... just so that it made sense.

> Remove use of INode.parent in Block CRC upgrade
> -----------------------------------------------
>
>                 Key: HADOOP-1774
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1774
>             Project: Hadoop
>          Issue Type: Improvement
>          Components: dfs
>            Reporter: Raghu Angadi
>            Assignee: Raghu Angadi
>             Fix For: 0.15.0
>
>         Attachments: HADOOP-1774.patch
>
>
> HADOOP-1687 proposes various reductions in Namenode memory. One of the 'cuts' is to remove 'parent' field in INodes (8 bytes per file). HADOOP-1743 already deprecates INode.getAbsoluteName() method. 
> Block CRC Upgrade code still requires this field (blockID -> fileName -> parent -> INode for .crc file). If we don't want to use parent, we need to build INode-to-parent map at the beginning of the upgrade. Note this would increase memory required during upgrade by 50 bytes for each of non-crc files (50% of files). I would estimate this to be around 5% more memory. 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.