hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Konstantin Shvachko (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HDFS-698) BlockSender should compare on disk file length with replica length instead of replica visible length.
Date Tue, 13 Oct 2009 01:17:31 GMT

    [ https://issues.apache.org/jira/browse/HDFS-698?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12764906#action_12764906
] 

Konstantin Shvachko commented on HDFS-698:
------------------------------------------

Member {{BlockSender.replica}} may be converted to a local variable in {{BlockSender}} constructor
because it is not used anywhere else.

> BlockSender should compare on disk file length with replica length instead of replica
visible length.
> -----------------------------------------------------------------------------------------------------
>
>                 Key: HDFS-698
>                 URL: https://issues.apache.org/jira/browse/HDFS-698
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: data-node
>    Affects Versions: 0.21.0
>            Reporter: Konstantin Shvachko
>             Fix For: 0.21.0
>
>
> BlockSender waits until all bytes of a replica are flushed to disk. Currently it compares
the replica bytes on disk with the replica visible length. In order to make sure that *all*
bytes are flushed it should compare the bytes on disk with the block length, because visible
length may be smaller than the block length.
> It would be useful also to create a special method FSDataset.getBytesOnDisk(blockID),
which gets this information through the volume map. Otherwise we may access replicas, which
are not in the map anymore.

-- 
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