hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Vinayakumar B (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HDFS-6115) flush() should be called for every append on block scan verification log
Date Tue, 18 Mar 2014 10:48:42 GMT

     [ https://issues.apache.org/jira/browse/HDFS-6115?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Vinayakumar B updated HDFS-6115:
--------------------------------

    Attachment: HDFS-6115.patch

Attached a simple patch to call out.flush().

Since the BlockScanning is throttled, calling out.flush() will not be big overhead on disk
I feel.

Please review

> flush() should be called for every append on block scan verification log
> ------------------------------------------------------------------------
>
>                 Key: HDFS-6115
>                 URL: https://issues.apache.org/jira/browse/HDFS-6115
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: datanode
>    Affects Versions: 2.3.0, 2.4.0
>            Reporter: Vinayakumar B
>            Assignee: Vinayakumar B
>         Attachments: HDFS-6115.patch
>
>
> {{RollingLogsImpl#out}} is  {{PrintWriter}} and it have the default buffersize of 8kB.
> So until the 8kB buffer is filled all scan verification entries will not be flushed,
hence there will be chance loosing this scan information if the datanode gets restarted. And
One more scan will happen for these blocks.
> 8kB will have ~80 entries, so 80 blocks needs to be re-scanned again.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message