[
https://issues.apache.org/jira/browse/PIG-1381?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12861344#action_12861344
]
V.V.Chaitanya Krishna commented on PIG-1381:
--------------------------------------------
bq. Do we need to have two different property files ? One possibility is to not
package pig.properties in the pig.jar and then include it in the classpath
while invoking Pig. (We can modify pig shell script to include it in the path
by default). Then, user can add/delete/modify the pig.properties as he wish as
well override default properties.
This might lead to overriding of some properties which might actually be
unacceptable. Also, in the long run, we might want to have a configuration file
with properties that are not supposed to be changed (similar to what happened
in case of Hadoop project)
bq. Disadvantage of two property files, is sometimes its confusing which
property is getting picked up (one in default or one in user specified). If
there is only one property file, there is only one way to specify the
properties to Pig which I think is better way of doing it.
Since the processing of properties' files is sequential (i.e., one file after
another), we can be sure that the latest occuring value is taken for a give
property. For example, we can load the default properties' file first and then
followed by the one in which users give their own set of properties. This way,
we could provide preference to users' settings as well.
Thoughts?
> Need a way for Pig to take an alternative property file
> -------------------------------------------------------
>
> Key: PIG-1381
> URL: https://issues.apache.org/jira/browse/PIG-1381
> Project: Pig
> Issue Type: Improvement
> Components: impl
> Affects Versions: 0.7.0
> Reporter: Daniel Dai
> Fix For: 0.8.0
>
>
> Currently, Pig read the first ever pig.properties in CLASSPATH. Pig has a
> default pig.properties and if user have a different pig.properties, there
> will be a conflict since we can only read one. There are couple of ways to
> solve it:
> 1. Give a command line option for user to pass an additional property file
> 2. Change the name for default pig.properties to pig-default.properties, and
> user can give a pig.properties to override
> 3. Further, can we consider to use pig-default.xml/pig-site.xml, which seems
> to be more natural for hadoop community. If so, we shall provide backward
> compatibility to also read pig.properties, pig-cluster-hadoop-site.xml.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
