|
I'm having issues with getting the JobTracker to stay alive. While trying to get Mahout working, I start of by running
However, when I start a JobTracker from the MapR Control System, it starts running. fs.JobTrackerWatcher is able to find it, but it ends up not being able to connect. Around the same time it starts to try connecting, it enters Some logs:
|
|
Do you run mahout from EC2 as well as your cluster? yes, everything is in the same security group on EC2; in the log above, both the JobTracker and Mahout are installed on 10.116.223.132;
(09 Apr '12, 18:31)
tristanls
Does JobTracker actually run? If yes, could you telnet to both ip-10-116-223-132.ec2.internal and 10.116.223.132 on port 9001 from where you are running mahout? I know it sounds strange, but your errors are also strange
(09 Apr '12, 22:47)
yufeldman ♦♦
|
|
Ok, I found
What I don't understand now, is why JAVA_HOME is not set? Because:
|
|
Ok, the "is JobTracker running?" problem is solved. My mistake was to start What is an issue is the mistery "Retrying connect to server" logs, however, they are followed by a "normal" failure, so this might be going well
My guess here is that Hadoop is looking for training data to be available on the JobTracker machine, instead of where the job originated from, i.e. the Mahout machine. |
|
Hadoop most likely is looking for input in maprfs where it is not there I guess. it looks to me like it's using $HOME instead of $MAHOUT_HOME because it's appending
(10 Apr '12, 14:46)
tristanls
|