Doug Balog 2011-05-12, 03:03
-Re: Question about hadoop namenode -format -clusterid <clusterID>
Correct way to format a namenode :
<hdfs_home>/bin/hdfs namenode -format -clusterid <cid>
PS: Set your environment right like common home etc.
Only first time it requires the cluster id, second time onwards it will remember cluster id and prompt you to format this particular cluster id.
I have filed a Jira on this:
From: Doug Balog <[EMAIL PROTECTED]>
To: [EMAIL PROTECTED]
Sent: Wednesday, May 11, 2011 8:03 PM
Subject: Question about hadoop namenode -format -clusterid <clusterID>
I'm at the hackathon in SF just trying to setup a single node cluster from my trunk checkout.
I'm at the point where I need to format a new namenode, and the old way of
just running "hadoop namenode -format" is failing because I'm not specifying a clusterID.
So I started poking around the code to try and figure what is expected for the clusterID
and I found that the namenode had a hidden option "-genclusterid" which
causes the namenode to just print out a new clusterID and exit.
I say hidden because if you run "hadoop namenode -usage" its not one of the listed
What is the correct way to format a namenode now (in trunk) ?
The current documentation doesn't match what the code does,
so its unclear to me how this is supposed to work.
IMHO "bin/namenode -format" should automaticlly generate a clusterID for you and
it should exit with an Exception. This is what everybody has been trained to do.
The only time you should have to specify a clusterID is when you want to
add a namenode to an existing cluster.