Hi Uri, The script start-mapred.sh has two commands one of them is used to start the jobtracker and the other is used to start the tasktrackers listed in the slaves file. I made a copy of the start-mapred.sh and removed the start job tracker command line. I change the slaves file according to what tasktarckers I want to start. I did the same with the stop-mapred.sh to stop the running tasktrackers.
It is a simple way to customize the start and stop tasktrackers. I think I will need to change the mapred script if I want to add the list of tasktrackers to start and stop in the command line. ________________________________ From: Uri Shani <[email protected]> To: [email protected] Sent: Friday, July 3, 2009 12:16:14 AM Subject: Re: starting a tasktracker on a specific node in the cluster So, what exactly did you do? From: Iman E <[email protected]> To: [email protected] Date: 03/07/2009 04:34 AM Subject: Re: starting a tasktracker on a specific node in the cluster The method I described below is now working! The jobtracker takes sometime to update its list of available task trackers (I think it is the timeout of a tasktracker) ________________________________ From: Iman E <[email protected]> To: [email protected] Sent: Thursday, July 2, 2009 4:18:10 PM Subject: starting a tasktracker on a specific node in the cluster Hi all, I want to start and stop tasktrackers one at a time without using the start-mapred script which also starts the jobtracker. I have tried to do "$HADOOP_CORE_HOME"/bin/hadoop-daemons.sh --config $HADOOP_CONF_DIR --script "$bin"/mapred start tasktracker but the tasktracker starts on the same node as the jobtracker which is not exactly what I want. I expected it to get information about where to start the task tracker from the slaves file. I have also tried to run this on the node where I want the tasktracker to start but it did not work either. Any pointers how to get this running? Thanks Iman
