You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-dev@hadoop.apache.org by "dhruba borthakur (JIRA)" <ji...@apache.org> on 2007/12/10 20:09:43 UTC
[jira] Issue Comment Edited: (HADOOP-2158) hdfsListDirectory in
libhdfs does not scale
[ https://issues.apache.org/jira/browse/HADOOP-2158?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12547224 ]
dhruba edited comment on HADOOP-2158 at 12/10/07 11:08 AM:
---------------------------------------------------------------------
Good patch. Code looks good. + 1.
II think the listPaths API on a directory used to return the size of the entire directory subtree. However, the listStatus API on a directory does not do so. If your application is not relying the original behaviour of listPaths then this change makes sense.
was (Author: dhruba):
Good patch. Code looks good. + 1.
II think the listPaths API on a directory used to return the size of the entire directory subtree. However, the listStatus API on a directory does not do so. If your application is not replying the original behaviour of listPaths then this change makes sense.
> hdfsListDirectory in libhdfs does not scale
> -------------------------------------------
>
> Key: HADOOP-2158
> URL: https://issues.apache.org/jira/browse/HADOOP-2158
> Project: Hadoop
> Issue Type: Bug
> Components: libhdfs
> Affects Versions: 0.15.0
> Reporter: Christian Kunz
> Assignee: Christian Kunz
> Priority: Blocker
> Fix For: 0.15.2
>
> Attachments: 2158.patch
>
>
> hdfsListDirectory makes one rpc call using deprecated fs.FileSystem.listPaths, and then two rpc calls for every entry in the returned array. When running a job with more than 3000 mappers each running a pipes application using libhdfs to scan a dfs directory with about 100-200 entries, this results in about 1M rpc calls to the namenode server overwhelming it.
> hdfsListDirectory should call fs.FileSystem.listStatus instead.
> I will submit a patch.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.