A new "ILLUSTRATE" command which will help people debug their pig programs
--------------------------------------------------------------------------
Key: PIG-59
URL: https://issues.apache.org/jira/browse/PIG-59
Project: Pig
Issue Type: New Feature
Components: grunt
Reporter: Shubham Chopra
I propose to add a new "ILLUSTRATE" command to Pig, which will help people
debug their Pig programs.
The idea is to select a few example data items, and illustrate how they are
transformed by the sequence of Pig commands in the user's program. I have an
algorithm that can select an appropriate and concise set of example data items
automatically. It does a better job than random sampling would do; for example,
random sampling suffers from the drawback that selective operations such as
filters or joins can eliminate *all* the sampled data items, giving you empty
results which is of no help in debugging.
This "ILLUSTRATE" functionality will avoid people having to test their Pig
programs on large data sets, which has a long turnaround time and wastes system
resources.
Proposed Implementation:
I will create a new package called org.apache.pig.exgen, which will contain the
aforementioned algorithm. The algorithm uses the "Local" execution operators
(it does not run on hadoop), so as to generate illustrative example data in
near-real-time for the user.
For my algorithm to work properly, it needs to trace the "lineage" (sometimes
called "provenance") of data items as they flow through the local operator tree
corresponding to the user's Pig program. So I will have to add a "lineage
tracer" to the Local operators, which maintains a side data structure to
represent the lineage, or derivation sequence, among data items. The lineage
tracer will be DISABLED BY DEFAULT, so it will not affect normal Pig operation.
I will add a new method to PigServer called
"PigServer.showExamples(LogicalPlan)", which will cause my exgen algorithm to
be invoked.
I will also add a new command to Grunt, called ILLUSTRATE. Syntactically it
will work the same way as the STORE command. For example, a user might type:
grunt> visits = load 'visits.txt' as (user, url, timestamp);
grunt> recent_visits = filter visits by timestamp >= '20071201';
grunt> user_visits = group recent_visits by user;
grunt> num_user_visits = foreach user_visits generate group,
COUNT(recent_visits);
grunt> illustrate num_user_visits
This would trigger my exgen algorithm, which will display something like:
visits:
(Amy, www.cnn.com, 20070218)
(Fred, www.harvard.edu, 20071204)
(Amy, www.bbc.com, 20071205)
(Fred, www.stanford.edu, 20071206)
recent_visits:
(Fred, www.harvard.edu, 20071204)
(Amy, www.bbc.com, 20071205)
(Fred, www.stanford.edu, 20071206)
user_visits:
(Fred, { (Fred, www.harvard.edu, 20071204), (Fred, www.stanford.edu, 20071206)
} )
(Amy, { (Amy, www.bbc.com, 20071205) } )
num_user_visits:
(Fred, 2)
(Amy, 1)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
