hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Konstantin Shvachko (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HADOOP-4358) NPE from CreateEditsLog
Date Wed, 15 Oct 2008 22:14:44 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-4358?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12639988#action_12639988
] 

Konstantin Shvachko commented on HADOOP-4358:
---------------------------------------------

Yes, the image stores the long value, that is we store the value returned by getAccessTime()
and the precision is factored out at this point. It is just that the value is rounded up to
the latest whole hour. So if we start a cluster with the new precision value then this rounded
value will be divided by the new precision and everything is correct. So, there is no issue
with configuring precision value as I suggested before, and Dhruba can safely run different
clusters with different precisions. Sorry about the confusion.
# The question is how to get rid of the static method call in INode.
#- I think we should change setAccessTime to {{setAccessTime(lonk aTime, int precision)}}
and propagate changes to the INode constructors.
#- Alternatively, we can create a {{static FSNmesystem.getPresision()}}, which would return
the default precision if {{FSNmesystem.this == null}}. Don't really like it, seems like another
shortcut to me.
# Second question is in what units the precision should be specified in configuration?
Aren't we introducing a Y2K-type bug even if the precision is specified it in hours?
# Raghu> _INode should be totally unaware of precision._
I don't think we can achieve this if we do not want to store the entire long for each file/directory
object.


> NPE from CreateEditsLog
> -----------------------
>
>                 Key: HADOOP-4358
>                 URL: https://issues.apache.org/jira/browse/HADOOP-4358
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: dfs, test
>    Affects Versions: 0.19.0
>            Reporter: Chris Douglas
>            Priority: Blocker
>             Fix For: 0.19.0
>
>         Attachments: HADOOP-4358.patch
>
>
> HADOOP-1869 added a call to setAccessTime(long) from the INode cstr, which relies on
a non-null value from FSNamesystem::getFSNamesystem.
> {noformat}
> java.lang.NullPointerException
>         at org.apache.hadoop.hdfs.server.namenode.INode.setAccessTime(INode.java:301)
>         at org.apache.hadoop.hdfs.server.namenode.INode.<init>(INode.java:99)
>         at org.apache.hadoop.hdfs.server.namenode.INodeDirectory.<init>(INodeDirectory.java:45)
>         at org.apache.hadoop.hdfs.CreateEditsLog.addFiles(CreateEditsLog.java:68)
>         at org.apache.hadoop.hdfs.CreateEditsLog.main(CreateEditsLog.java:214)
> {noformat}

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


Mime
View raw message