JobTracker assigns TaskTracker role to a server of the cluster that it shoud not use...

[Björn-Elmar Macek <em...@cs.uni-kassel.de>]

Hi all,

well since i now got all servers continiously running for my job i still 
encounter problems:

although all services seem to be up and no errors are produced i seem to 
be stuck at the map process at a certain percentage. I am not yet sure, 
that just letting the cluster run may solve the issue, but somehow it 
seems wrong, since mapping up to 62% is done nearly instantly (logs are 
growing coniniously) and from that point it gets really slow.

The only hint i see is the following in the jobtracker log is, that it 
says that there is *serious* trouble with some task tracker and that it 
takes a new one... but somehow it takes a server that was not configured 
or assigned to any task, since it doesnt occur in ANY of the config files.

2012-08-22 13:55:31,181 WARN org.apache.hadoop.mapred.JobTracker: 
Serious problem, cannot find record of 'previous' heartbeat for 
'tracker_its-cs213.its.uni-kassel.de:localhost/127.0.0.1:51847'; 
reinitializing the tasktracker
2012-08-22 13:55:34,217 INFO org.apache.hadoop.net.NetworkTopology: 
Adding a new node: /default-rack/its-cs213.its.uni-kassel.de
2012-08-22 13:55:34,218 INFO org.apache.hadoop.mapred.JobTracker: Adding 
tracker tracker_its-cs213.its.uni-kassel.de:localhost/127.0.0.1:39168 to 
host its-cs213.its.uni-kassel.de

What am i experiencing here? Is there any way to forbid hadoop to take 
new nodes (because they do not belong to me and may be intensively used 
by other users for calculation)?

Regards,
Elmar