Check out how does the writing happens on HDFS...
When client issues the command "hadoop fs -put local_source
hdfs_destination", client contact the namenode that he wants to write and
then NameNode creates a blockID and ask three datanodes ( if replication on
the client side is set to 3) to host the replica. THis information is sent
to client and then client starts writing the data on to the data nodes by
forming pipe line..And then client write that much amount of data onto a
block which he has set.
Som Shekhar Sharma
On Sun, Jul 14, 2013 at 4:24 PM, Harsh J <[EMAIL PROTECTED]> wrote:
> Replication, block size, etc. are all per-file and pure client
> supplied properties. They either take their default from the client
> config, or directly from an API argument override.
> On Sun, Jul 14, 2013 at 4:14 PM, varun kumar <[EMAIL PROTECTED]> wrote:
> > What Shumin told is correct,hadoop configurations has been over written
> > through client application.
> > We have faced similar type of issue,Where default replication factor was
> > mentioned 2 in hadoop configuration.But when when ever the client
> > application writes a files,it was having 3 copies in hadoop
> cluster.Later on
> > checking client application it's default replica factor has 3.
> > On Sun, Jul 14, 2013 at 4:51 AM, Shumin Guo <[EMAIL PROTECTED]> wrote:
> >> I Think the client side configuration will take effect.
> >> Shumin
> >> On Jul 12, 2013 11:50 AM, "Shalish VJ" <[EMAIL PROTECTED]> wrote:
> >>> Hi,
> >>> Suppose block size set in configuration file at client side is
> >>> block size set in configuration file at name node side is 128MB and
> >>> size set in configuration file at datanode side is something else.
> >>> Please advice, If the client is writing a file to hdfs,which property
> >>> would be executed.
> >>> Thanks,
> >>> Shalish.
> > --
> > Regards,
> > Varun Kumar.P
> Harsh J