TriHUG November Pig Talk by Alan Gates
- 1. Pig Dataflow Scripting for Hadoop Alan F. Gates @alanfgates © Hortonworks, Inc 2011 Page 1
- 2. Who Am I? • Pig committer and PMC Member • HCatalog committer and mentor • Member of ASF and Incubator PMC • Co-founder of Hortonworks • Author of Programming Pig from O’Reilly Photo credit: Steven Guarnaccia, The Three Little Pigs
- 3. Who Are You? 3
- 4. Example For all of your Load Users Load Logs registered users, you Semi-join want to count how many came to your site Count by zip Count by this month. You want age, gender this count both by geography (zip code) Store Store results results and by demographic group (age and gender)
- 5. In Pig Latin -- Load web server logs logs = load 'server_logs' using HCatLoader(); thismonth = filter logs by date >= '20110801' and date < '20110901'; -- Load users users = load 'users' using HCatLoader(); -- Remove any users that did not visit this month grpd = cogroup thismonth by userid, users by userid; fltrd = filter grpd by not IsEmpty(logs); visited = foreach fltrd generate flatten(users); -- Count by zip code grpbyzip = group visited by zip; cntzip = foreach grpbyzip generate group, COUNT(visited); store cntzip into 'by_zip' using HCatStorer('date=201108'); -- Count by demographics grpbydemo = group visited by (age, gender); cntdemo = foreach grpbydemo generate flatten(group), COUNT(visited); store cntdemo into 'by_demo' using HCatStorer('date=201108');
- 6. Pig’s Place in the Data World Data Collection Data Factory Data Warehouse Pig Hive Pipelines BI Tools Iterative Processing Analysis Research 6
- 7. Why not MapReduce? • Pig Provides a number of standard data operators – Five different implementations of join (hash, fragment- replicate, merge, sparse merged, skewed) – Order by provides total ordering across reducers in a balanced way • Provides optimizations that are hard to do by hand – Multi-query: Pig will combine certain types of operations together in a single pipeline to reduce the number of times data is scanned • User Defined Functions provide a way to inject your code into the data transformation – can be written in Java or Python – can do column transformation (TOUPPER) and aggregation (SUM) – can be written to take advantage of the combiner • Control flow can be done via Python or Java 7
- 8. Embedding Example: Compute Pagerank PageRank: A system of linear equations (as many as there are pages on the web, yeah, a lot): It can be approximated iteratively: compute the new page rank based on the page ranks of the previous iteration. Start with some value. Ref: http://en.wikipedia.org/wiki/PageRank Slide courtesy of Julien Le Dem
- 9. Or more visually Each page sends a fraction of its PageRank to the pages linked to. Inversely proportional to the number of links. Slide courtesy of Julien Le Dem
- 10. Slide courtesy of Julien Le Dem
- 11. Let’s zoom in pig script: PR(A) = (1-d) + d (PR(T1)/C(T1) + ... + PR(Tn)/C(Tn)) Iterate 10 times Pass parameters as a dictionary Just run P, that was declared above The output becomes the new input Slide courtesy of Julien Le Dem
- 12. Recently Added Features • New in 0.9 (released July 2011): – Embedding in Python – Macros and Imports • New in 0.10 (should release in Dec 2011) – Boolean data type – Hash based aggregation for aggregates with low cardinality keys – UDFs to build and apply bloom filters – UDFs in JRuby (may slip to next release) 14
- 13. Learn More • Read the online documentation: http://pig.apache.org/ • Programming Pig from O’Reilly Press • Join the mailing lists: – [email protected] for user questions – [email protected] for developer issues • Follow me on Twitter, @alanfgates
- 14. Questions 16
Editor's Notes
- #3: Say a little about Hortonworks
- #7: SQL is a query languageDeclarative, what not howOriented around answering a questionRequires uniform schemaRequires metadataKnown by everyoneA great choice for answering queries, building reports, use with automated toolsPig Latin is a data flow languageScript defines a data flowIntended for pipelines where there may be tens or hundreds of stepsBuilt for raw world of Hadoop where schemas are optional, data may not be clean, etc.Can operate with or without metadataA great choice for ETL pipelines, data models, iterative processing, and research on raw data