i thought that start-all and stop-all weren't supposed to be used on distributed clusters... that it was just sugar for testing/learning...
try start-hdfs then start-mapred (and if you stop them, it's stop-mapred then stop-hdfs) then again it's ec2 so they might do something weird or special that i'm not aware of. -mike On Tue, Nov 24, 2009 at 1:24 PM, Stephen Watt <[email protected]> wrote: > Hi Mark > > Are you starting the clusters from the contrib/ec2 scripts ? These scripts > have a special way of bringing up the cluster where they are passing in > the hostnames of the slaves as they are being assigned from ec2, thus I > think stop-all and start-all will not work as they both assume the slaves > are defined in the slaves file. Its been awhile since I looked at this so > excuse my lack of specifics. I believe there is a script in the /root > directory of each ec2 image that these values are being passed into that > does the work of starting the tasktracker/datanode processes on each one > of these. > > Kind regards > Steve Watt > > > > From: > Mark Kerzner <[email protected]> > To: > [email protected] > Date: > 11/24/2009 03:02 PM > Subject: > Hadoop on EC2 > > > > Hi, > > I am starting a cluster of Apache Hadoop distributions, like .18 and also > .19. This all works fine, then I log in. I see that the Hadoop daemons are > already working. However, when I try > > # which hadoop > /usr/local/hadoop-0.19.0/bin/hadoop > # jps > 1355 Jps > 1167 NameNode > 1213 JobTracker > # hadoop fs -ls hdfs://localhost/ > 09/11/24 15:33:56 INFO ipc.Client: Retrying connect to server: localhost/ > 127.0.0.1:8020. Already tried 0 time(s). > > I do stop-all.sh and then start-all.sh, and it does not help. What am I > doing wrong? > > Thank you, > Mark > > >
