That's bizarre. I'm not sure why your DFS would have magically gotten
full. Whenever hadoop gives me trouble, i try the following sequence
of commands
stop-all.sh
rm -Rf /path/to/my/hadoop/dfs/data
hadoop namenode -format
start-all.sh
maybe you would get some luck if you ran that on all of the machines?
(of course, don't run it if you don't want to lose all of that "data")
On Jun 19, 2008, at 4:32 AM, novice user wrote:
Hi Every one,
I am running a simple map-red application similar to k-means. But,
when I
ran it in on single machine, it went fine with out any issues. But,
when I
ran the same on a hadoop cluster of 9 machines. It fails saying
java.io.IOException: All datanodes are bad. Aborting...
Here is more explanation about the problem:
I tried to upgrade my hadoop cluster to hadoop-17. During this
process, I
made a mistake of not installing hadoop on all machines. So, the
upgrade
failed. Nor I was able to roll back. So, I re-formatted the name node
afresh. and then hadoop installation was successful.
Later, when I ran my map-reduce job, it ran successfully,but the
same job
with zero reduce tasks is failing with the error as:
java.io.IOException: All datanodes are bad. Aborting...
When I looked into the data nodes, I figured out that file system is
100%
full with different directories of name "subdir" in
hadoop-username/dfs/data/current directory. I am wondering where I
went
wrong.
Can some one please help me on this?
The same job went fine on a single machine with same amount of input
data.
Thanks
--
View this message in context:
http://www.nabble.com/java.io.IOException%3A-All-datanodes-are-bad.-Aborting...-tp18006296p18006296.html
Sent from the Hadoop core-user mailing list archive at Nabble.com.
