Hi,

I was just wondering why do we specify more than one directories for
dfs.name.dir
.

What I understand is dfs.name.dir should be valued to 2 directories one on
the local and the other on the NFS.

 Are we trying to protect the name node from eventualities?  If so, what is
the role of Secondary NameNode ?

I am using apache hadoop-0.20

Regards,
Nagarjuna

Hi Nagarjuna

Multiple dfs.name.dir is for more reliability so that you'll be having multiple copies of namespace image.

Secondary name node does the checkpointing, ie merges the fsimage and edit log at defined periodic intervals. So SecondaryNameNode at a time will have the most recent copy of fs image but no current edit log. As the NameNode would already have a new edit log after the previous one was passed to SNN for merging.
 
Regards
Bejoy KS

Sent from handheld, please excuse typos.

-----Original Message-----
From: nagarjuna kanamarlapudi <[EMAIL PROTECTED]>
Date: Sat, 17 Nov 2012 18:40:37
To: <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Subject: Multiple directories for dfs.name.dir

Hi,

I was just wondering why do we specify more than one directories for
dfs.name.dir
.

What I understand is dfs.name.dir should be valued to 2 directories one on
the local and the other on the NFS.

 Are we trying to protect the name node from eventualities?  If so, what is
the role of Secondary NameNode ?

I am using apache hadoop-0.20

Regards,
Nagarjuna