Andy, many thanks. I am stuck here now so please put me in the right direction.
I successfully ran a job on a cluster on foo1 in pseudo-distributed mode and are now trying to try fully-dist'ed one. a. I created another instance foo2 on EC2. Installed hadoop on it and copied conf/ folder from foo1 to foo2. I created /hadoop/dfs/data folder on the local linux system on foo2. b. on foo1 I created file conf/slaves and added: localhost <hostname-of-foo2> At this point I cannot find an answer on what to do next. I started NN, DN, SNN, JT, TT on foor1. After I ran "hadoop fsck /user/bar -files -blocks -locations", it showed # of datanode as 1. I was expecting DN and TT on foo2 to be started by foo1. But it didn’t happen, so I started them myself and tried the the command again. Still one DD. I realise that boo2 has no data at this point but I could not find bin/start-balancer.sh script to help me to balance data over to DD from foo1 to foo2. What do I do next? Thanks AK -----Original Message----- From: Andy Isaacson [mailto:[email protected]] Sent: Friday, October 26, 2012 2:21 PM To: [email protected] Subject: Re: cluster set-up / a few quick questions On Fri, Oct 26, 2012 at 9:40 AM, Kartashov, Andy <[email protected]> wrote: > Gents, We're not all male here. :) I prefer "Hadoopers" or "hi all,". > 1. > - do you put Master's node <hostname> under fs.default.name in core-site.xml > on the slave machines or slaves' hostnames? Master. I have a 4-node cluster, named foo1 - foo4. My fs.default.name is hdfs://foo1.domain.com. > - do you need to run "sudo -u hdfs hadoop namenode -format" and create /tmp > /var folders on the HDFS of the slave machines that will be running only DN > and TT or not? Do you still need to create hadoop/dfs/name folder on the > slaves? (The following is the simple answer, for non-HA non-federated HDFS. You'll want to get the simple example working before trying the complicated ones.) No. A cluster has one namenode, running on the machine known as the master, and the admin must "hadoop namenode -format" on that machine only. In my example, I ran "hadoop namenode -format" on foo1. > 2. > In hdfs-site.xml for dfs.name.dir & dfs.data.dir properties we specify > /hadoop/dfs/name /hadoop/dfs/data being local linux NFS directories by > running command "mkdir -p /hadoop/dfs/data" > but mapred.system.dir property is to point to HDFS and not NFS since we are > running "sudo -u hdfs hadoop fs -mkdir /tmp/mapred/system"?? > If so and since it is exactly the same format /far/boo/baz how does hadoop > know which directory is local on NFS or HDFS? This is very confusing, to be sure! There are a few places where paths are implicitly known to be on HDFS rather than a Linux filesystem path. mapred.system.dir is one of those. This does mean that given a string that starts with "/tmp/" you can't necessarily know whether it's a Linux path or a HDFS path without looking at the larger context. In the case of mapred.system.dir, the docs are the place to check; according to cluster_setup.html, mapred.system.dir is "Path on the HDFS where where the Map/Reduce framework stores system files". http://hadoop.apache.org/docs/r1.0.3/cluster_setup.html Hope this helps, -andy NOTICE: This e-mail message and any attachments are confidential, subject to copyright and may be privileged. Any unauthorized use, copying or disclosure is prohibited. If you are not the intended recipient, please delete and contact the sender immediately. Please consider the environment before printing this e-mail. AVIS : le présent courriel et toute pièce jointe qui l'accompagne sont confidentiels, protégés par le droit d'auteur et peuvent être couverts par le secret professionnel. Toute utilisation, copie ou divulgation non autorisée est interdite. Si vous n'êtes pas le destinataire prévu de ce courriel, supprimez-le et contactez immédiatement l'expéditeur. Veuillez penser à l'environnement avant d'imprimer le présent courriel
