hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Devaraj Das (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HADOOP-6843) Entries in the FileSystem's Cache could be cleared when they are not used
Date Sat, 26 Jun 2010 04:18:51 GMT

     [ https://issues.apache.org/jira/browse/HADOOP-6843?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Devaraj Das updated HADOOP-6843:

    Attachment: fs-weak-ref.4.patch

An approach to the above problem is to make the value in the map (used to implement the cache)
a soft reference. Then when no one is referencing a particular FileSystem instance, and the
memory is under pressure, the GC can delete the entry (and unless the Configuration object
is referenced from elsewhere, that would go away too). 

Patch for Y20S (not for commit).

> Entries in the FileSystem's Cache could be cleared when they are not used
> -------------------------------------------------------------------------
>                 Key: HADOOP-6843
>                 URL: https://issues.apache.org/jira/browse/HADOOP-6843
>             Project: Hadoop Common
>          Issue Type: Bug
>            Reporter: Devaraj Das
>             Fix For: 0.22.0
>         Attachments: fs-weak-ref.4.patch
> In FileSystem, there is a cache maintained for Filesystem instances. The entries in the
cache are cleared only when explicit FileSystem.close is invoked. Applications are not careful
on this aspect. Typically, they do FileSystem.get(), operate on the FileSystem, and then they
just forget about it. Every FileSystem instance stores a reference to the Configuration object
that it was created with. Over a period of time, as the cache grows, this can lead to OOM
(we have seen this happening in our hadoop 20S clusters at Yahoo).
> This jira aims at addressing the above issue.

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

View raw message