Thanks for the suggestion. I'll give it a try. First look is PIG-2183 doesn't apply cleanly to 0.9, but it does seem to work against trunk. I'll look further but I was more looking at a solution that keeps me from re-building the jar from source to avoid running into the situation referenced in PIG-2055.
-----Original Message----- From: Daniel Dai [mailto:[email protected]] Sent: Thursday, August 04, 2011 4:24 PM To: [email protected] Subject: Re: using specific hadoop version with Pig 0.9... Can you try https://issues.apache.org/jira/browse/PIG-2183? Do a "ant pig-withouthadoop" first, then run bin/pig. Daniel On Thu, Aug 4, 2011 at 12:56 PM, Andy Sautins <[email protected]> wrote: > > I have been very excited to give Pig 0.9 a try and run it against our > Cloudera CDH3U0 hadoop cluster and I need to point Pig to the cloudera hadoop > libraries to make it work. I tried re-building pig and ran into the behavior > described in PIG-2055. My other option seems to be to set the > PIG_HADOOP_VERSION environment variable and add the cloudera hadoop-core jar > to the pig lib directory. This did not seem to work, however if I change > bin/pig to put the inclusion of the hadoop jar in front of the pig jar in the > classpath it works as I would expect ( i.e., it includes the Cloudera > hadoop-core jar before pig-core.jar and runs correctly. ) > > The question I have is does this approach seem to make sense? If so does > it seem worthwhile to log a JIRA or am I just mis-understanding how to > specify an alternate hadoop distribution. It does look like bin/pig in trunk > is different although it also seems to include pig-core.jar before the hadoop > jars in the classpath. > > Thanks > > >
