stack-3 wrote:
>
>
>>> Running a MR
>>>job, participating classes must be available on the CLASSPATH on all
nodes,
>>>not just client.
>>>
>>>My guess is that running from eclipse you were not fulfilling the latter
>>>requirement.
>>>
>>>Did you try running the rowcounter program from the command-line?
>>>
>>>St.Ack
>
>
Hi St.Ack,
The purpose of adding the HBase jar to the Hadoop classpath export was to
make HBase class availabliity known and usable to Hadoop. That of course is
on the Linux machine where both Hadoop and HBase are running. That was the
means supposed to fulfill the HBase classes availablity you mention above.
It did not seem to work. But, physically copying the HBase jar over to the
Hadoop deployment DID work. Again, on the Linux machine, not on the windows
client machine. Running a client process on a remote windows machine did not
generate the error in any way, or it would not have been fixed by the
physical jar copy to Hadoop on the server machine. And that did fix it. So
the real question I see is why did the change to the HADOOP_CLASSPATH
export, on the Hadoop deployment, NOT work as it was supposed to? That is
maybe the thing which needs to be looked at.
No, I did not try running the rowcounter code.
Thanks for the response, and I hope this makes the issue clearer.
(So, right now I am applying the 660 shim patches to PIG 0.3.0 - hoping to
get that running tomorrow. )
_mc
--
View this message in context:
http://www.nabble.com/Exception-in-rowcount-program-tp23195921p24957698.html
Sent from the HBase User mailing list archive at Nabble.com.
