jay vyas created BIGTOP-1089:
--------------------------------
Summary: BigPetStore: A bigtop blueprint project inside of bigtop
Key: BIGTOP-1089
URL: https://issues.apache.org/jira/browse/BIGTOP-1089
Project: Bigtop
Issue Type: New Feature
Components: General
Reporter: jay vyas
The need for templates for processing big data pipelines is obvious - and also
- given the increasing amount of overlap across different big data and nosql
projects, it will provide a ground truth in the future for comparing the
behaviour and approach of different tools to solve a common, easily
comprehended problem.
This ticket formalizes the conversation in mailing list archives regarding the
BigPetStore proposal.
At the moment, (with the exception of word count), there are very few examples
of bigdata problems that have been solved by a variety of different
technologies. And, even with wordcount, there arent alot of templates which
can be customized for applications.
Comparatively: Other application developer communities (i.e.the Rails folks,
those using maven archetypes, etc.. ) have a plethora of template applications
which can be used to kickstart their applications and use cases.
This big pet store JIRA thus aims to do the following:
0) Curate a single, central, standard input data set .
1) Define a big data processing pipeline (using the pet store theme - except
morphing it to be analytics rather than transaction oriented).
2) Define specific nodes as DAG in the pipeline. For example, the first step
would be to ETL some text files into the Hadoop cluster's default FileSystem.
Each task would have multiple implementations, using different component of the
bigtop stack. Using word count as the example: We would have word count
implementations in pig, hive, mapreduce, spark, etc... and unit tests which
confirmed that each one accomplished the exact same output.
3) Package the project as a maven archetype that can be used to create new java
based bigdata processor frameworks.
Some implementation details -- open to change these, please comment/review -- .
- initial data source will be raw text or (better yet) some kind of
automatically generated data.
- the source will initially go in bigtop/blueprints
- the application sources can be in any modern JVM language
(java,scala,groovy,clojure), since bigtop supports scala, java, groovy natively
already and clojure is easy to support with the right jars.
- each "job" will be named according to the corresponding DAG of the big data
pipeline .
- all jobs will be controlled by a global program (maybe oozie?) which runs the
tasks in order, and can easily be customized to use different tools at
different stages.
- for now, all outputs will be to files: so that users don't require servers to
run the app.
- final data sinks will be into a highly available transaction oriented store
(solr/hbase/riak/...)
This ticket will be completed once a first iteration of BigPetStore is complete
using 3 ecosystem components, along with a depiction of the pipeline which can
be used for development.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
