hadoop-mapreduce-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rahul Chaudhari <rahulchaudhari0...@gmail.com>
Subject Re: full partition handling
Date Thu, 29 May 2014 19:06:58 GMT
There is no error reported as that specific partition will not be used for
further write requests. In the configurations, you can specify multiple
disk partitions for data node and even if any of that partition does not
exists, it is simply skipped by hadoop.

_Rahul


On Thu, May 29, 2014 at 8:39 PM, Aitor Perez Cedres <aperez@pragsis.com>wrote:

>
> Hi Brian,
>
> I'm not sure how it will react; I guess the datanode will report an error
> of no space left on device and fail to serve some requests trying to write
> on the partition that is full.
>
> To solve this issue, you have to manually move files from
> *dfs.datanode.data.dir*/current to a different partition. Also you have
> to stop the datanode before this "re-balancing".
>
>
> On 29/05/14 16:52, Brian Jeltema wrote:
>
> Google wasn’t helpful for this question:
>
>  If a Hadoop cluster has multiple disk partitions assigned to HDFS and one of those partitions
is full,
>  how does HDFS react?
>
> TIA
>
> Brian
>
>
> --
> *Aitor Pérez*
> *Big Data System Engineer*
>
> Telf.: +34 917 680 490
> Fax: +34 913 833 301
> C/Manuel Tovar, 49-53 - 28034 Madrid - Spain
>
> *http://www.bidoop.es <http://www.bidoop.es>*
>



-- 
Regards,
Rahul Chaudhari

Mime
View raw message