Data in HDFS is read and written via the individual DN's 50010 ports, which you would also need to open up to avoid these errors. Data isn't written/read through the NameNode.
On Thu, Nov 7, 2013 at 4:50 AM, Sergey Gerasimov <[email protected]> wrote: > Hello, > > > > I have problems with posting jar to my cluster remotely from client machine > located somewhere in the Web. I use original hadoop-1.2.1. > > > > I installed hadoop on client machine (same version as in the cluster), > configured fs.default.name and mapred.job.tracker. > > Access to DFS works fine remotely. I can successfully play with “hadoop fs” > commands. > > > > But when I send some job, for example: > > hadoop jar hadoop-examples -1.2.1.jar sleep 1 > > > > I see output like: > > 13/11/07 02:44:42 INFO hdfs.DFSClient: Exception in createBlockOutputStream > xx.xx.xx.xx:50010 java.net.ConnectException: Connection timed out > > 13/11/07 02:44:42 INFO hdfs.DFSClient: Abandoning > blk_1089181243677159149_31717 > > 13/11/07 02:44:42 INFO hdfs.DFSClient: Excluding datanode xx.xx.xx.xx:50010 > > 13/11/07 02:45:45 INFO hdfs.DFSClient: Exception in createBlockOutputStream > xx.xx.xx.xx:50010 java.net.ConnectException: Connection timed out > > 13/11/07 02:45:45 INFO hdfs.DFSClient: Abandoning > blk_6550586867464091073_31717 > > 13/11/07 02:45:45 INFO hdfs.DFSClient: Excluding datanode xx.xx.xx.xx:50010 > > 13/11/07 02:46:48 INFO hdfs.DFSClient: Exception in createBlockOutputStream > xx.xx.xx.xx:50010 java.net.ConnectException: Connection timed out > > 13/11/07 02:46:48 INFO hdfs.DFSClient: Abandoning > blk_5814098597599107248_31717 > > 13/11/07 02:46:48 INFO hdfs.DFSClient: Excluding datanode xx.xx.xx.xx:50010 > > 13/11/07 02:47:51 INFO hdfs.DFSClient: Exception in createBlockOutputStream > xx.xx.xx.xx:50010 java.net.ConnectException: Connection timed out > > 13/11/07 02:47:51 INFO hdfs.DFSClient: Abandoning > blk_6368219524592897749_31717 > > > > The same jar sent from inside the cluster runs fine. > > > > The network where cluster lives is protected by firewall with only NameNode > and JobTracker ports opened externally. > > iptables on all nodes are off. > > > > I have no ideas about reasons of these messages in the log. To the moment I > were sure that entry point to hadoop cluster contains just NameNode and > JobTracker ports. > > Both are open. > > > > Please help! > > > > > > -- Harsh J
