hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "dhruba borthakur (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HADOOP-893) dead datanode set should be maintained in the file handle or file system for hdfs
Date Mon, 05 Feb 2007 22:26:07 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-893?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12470381
] 

dhruba borthakur commented on HADOOP-893:
-----------------------------------------

+1. Code looks good.

The patch re-determines alive/dead datanodes only when all the replicas of a block reside
on dead-datanodes. Suppose an application opened a file and kept the handle active for an
extended period of time (when datanodes can go down and come back up). The randomization behaviour
of selecting a random remote node is now going to be different from earlier.

> dead datanode set should be maintained in the file handle or file system for hdfs
> ---------------------------------------------------------------------------------
>
>                 Key: HADOOP-893
>                 URL: https://issues.apache.org/jira/browse/HADOOP-893
>             Project: Hadoop
>          Issue Type: Bug
>          Components: dfs
>            Reporter: Owen O'Malley
>         Assigned To: Raghu Angadi
>         Attachments: HADOOP-893-1.patch, HADOOP-893-2.patch
>
>
> Currently each call to read is creating a new set of dead datanodes. It seems like it
would be more useful to keep a set of dead datanodes at either the file handle or file system
level since dead datanodes are probably not quite that transient.

-- 
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