How can I load external files within jython UDF?

[Young Ng <yo...@gmail.com>]

Hi,

I am trying to load some external resources within my jython udf functions, e.g:

@outputSchema(....)
def test():
    f = open('test.txt.')
    text = f.read()
    f.close()
    return text

I have place the 'test.txt' on both working folder and hdfs, and I got the following error:
   IOError: (2, 'No such file or directory', 'test.txt')

I have also try to print out the working path of jython, with os.getcwd(), below is what I got:
  /home/hduser/tmp/mapred/local/taskTracker/hduser/jobcache/job_201212080111_0007/attempt_201212080111_0007_m_000000_0/work
  ....

I suspect that I can use absolute path within udf, but how can I transfer the external resources to 
other hadoop datanodes?


Thanks,
Young Wu

Re: How can I load external files within jython UDF?

[Young Ng <yo...@gmail.com>]

Thank you John,

I have checked the SHIP function, but It's a bit confusing to me. all the examples I found were related to streaming-through. 
And I have figured out a solution by putting the file on hdfs, and specifying the following options when executing pig: 

pig -Dmapred.cache.files=hdfs://host:port/path/to/file#link_name -Dmapred.create.symlink=yes some.pig

then I can just use `f = open('link_name')` in my jython UDF. since the file is small, and loaded only once, it works good so far.

On Dec 9, 2012, at 2:18 PM, John Gordon <Jo...@microsoft.com> wrote:

> If you ship the file explicitly, you can use this syntax from there.  It will pack it with the job jar and make sure it is in the working directory wherever the job runs.  Be careful of shipping very large files, it is probably better to refactor your logic into multiple tiplevel pig statements on data loaded from hdfs if you find yourself shipping fixed, very large files.
> ________________________________
> From: Young Ng
> Sent: 12/9/2012 12:53 PM
> To: user@pig.apache.org
> Subject: How can I load external files within jython UDF?
> 
> Hi,
> 
> I am trying to load some external resources within my jython udf functions, e.g:
> 
> @outputSchema(....)
> def test():
>    f = open('test.txt.')
>    text = f.read()
>    f.close()
>    return text
> 
> I have place the 'test.txt' on both working folder and hdfs, and I got the following error:
>   IOError: (2, 'No such file or directory', 'test.txt')
> 
> I have also try to print out the working path of jython, with os.getcwd(), below is what I got:
>  /home/hduser/tmp/mapred/local/taskTracker/hduser/jobcache/job_201212080111_0007/attempt_201212080111_0007_m_000000_0/work
>  ....
> 
> I suspect that I can use absolute path within udf, but how can I transfer the external resources to
> other hadoop datanodes?
> 
> 
> Thanks,
> Young Wu
> 
> 
> 
> 
> 

RE: How can I load external files within jython UDF?

[John Gordon <Jo...@microsoft.com>]

If you ship the file explicitly, you can use this syntax from there.  It will pack it with the job jar and make sure it is in the working directory wherever the job runs.  Be careful of shipping very large files, it is probably better to refactor your logic into multiple tiplevel pig statements on data loaded from hdfs if you find yourself shipping fixed, very large files.
________________________________
From: Young Ng
Sent: 12/9/2012 12:53 PM
To: user@pig.apache.org
Subject: How can I load external files within jython UDF?

Hi,

I am trying to load some external resources within my jython udf functions, e.g:

@outputSchema(....)
def test():
    f = open('test.txt.')
    text = f.read()
    f.close()
    return text

I have place the 'test.txt' on both working folder and hdfs, and I got the following error:
   IOError: (2, 'No such file or directory', 'test.txt')

I have also try to print out the working path of jython, with os.getcwd(), below is what I got:
  /home/hduser/tmp/mapred/local/taskTracker/hduser/jobcache/job_201212080111_0007/attempt_201212080111_0007_m_000000_0/work
  ....

I suspect that I can use absolute path within udf, but how can I transfer the external resources to
other hadoop datanodes?


Thanks,
Young Wu