在Redshift中创建python UDF时出错

Question

I am trying to create Python UDF in Amazon Redshift, and I have successfully created the UDF with no error. I have also created the required library for this UDF successfully. But when I execute the UDF, I get the error:

No Module Named pyffx. Please look at svl_udf_log for more information

I have downloaded the library from pypi.org and uploaded it to Amazon S3. This is the link I used to download the library:

https://pypi.org/project/pyffx/#files

create library pyffx
language plpythonu
from 's3://aws-bucket/tmp/python_module/pyffx-0.3.0.zip'
credentials
'aws_iam_role=iam role'
region 'us-east-1';

CREATE OR REPLACE FUNCTION schema.ffx(src VARCHAR)
RETURNS VARCHAR
STABLE
AS $$
    import pyffx
    src = unicode(src)
    value=(src)
    l=len(value)
    e = pyffx.String(b'secret-key', alphabet='abcedefghijklmnopqrstuvwxyz123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ', length=l)
    return e.encrypt(value)
$$ LANGUAGE plpythonu;

Answer 1

I managed to get it to work... sort of.

I did the following:

• Downloaded pyffx via the link you provided
• Extracted the .tar.gz file and created a .zip of the files
• Copied the .zip file to Amazon S3
• Loaded the library using your CREATE LIBRARY command
• Created the function

However, when I use the function, I receive the error:

Invalid operation: AttributeError: 'module' object has no attribute 'add_metaclass'

My research suggests that the six library (that provides Python 2 and 3 compatibility) is the source of this problem. The Python Language Support for UDFs - Amazon Redshift page indicates that six 1.3 is included in Redshift, yet Pip six.add_metaclass error says that this version does not include add_metaclass . The current version of six is 1.12.

I tried to include an updated six library in the code but wasn't successful. You might be able to wrangle better than me.