reduce io during sharing scans of the same input datasets
----------------------------------------------------------
Key: PIG-614
URL: https://issues.apache.org/jira/browse/PIG-614
Project: Pig
Issue Type: Improvement
Components: impl
Affects Versions: types_branch
Reporter: Samuel Guo
Priority: Minor
Fix For: types_branch
If we want to store different results that generated from the same input
dataset, now we need to write two or several *STORE* clauses. And these *STORE*
clauses will be translated to different mr jobs despite of these mr jobs may
share scans of the same input datasets.
for example:
Dataset 'weather' contains the records of the weather. Each record contains
three part : wind/air/tempreture. we need to process different part of the
records.
we may write a pig script as below:
weather = load 'weather.txt' as (wind, air, tempreture);
wind_results = ... wind ...;
air_results = ...air...;
temp_results = ...tempreture...;
store wind_results into 'wind.results';
store air_results into 'air.results';
store temp_results into 'temp.results';
now pig will translate this script into three different MR jobs wich run
sequencely: scan 'weather.txt', process the wind data, store the wind results;
scan 'weather.txt' again, process the air data, store the air results; ...
if the input data set is large, it is not efficient.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.