-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/19633/
-----------------------------------------------------------

Review request for pig, Daniel Dai and Rohini Palaniswamy.


Bugs: PIG-3743
    https://issues.apache.org/jira/browse/PIG-3743


Repository: pig-git


Description
-------

The patch reimplements union using Tez VertexGroup and GroupInputEdge.

The changes include-
* Implemented POVertexGroupInputTez that takes ConcatenatedMergedKeyValuesInput 
from VertexGroup.
* TezCompiler inserts an alias vertex for union, and the alias vertex is 
converted to VertexGroup by TezDagBuilder.
* TezStats JobGraphBuilder removes alias vertices since they're not 
materialized by Tez, and thus, there is no status for them.

Note that-
* Further optimization is possible for the case where union is only followed by 
store. In that case, we could directly attach a MROutput to VertexGroup instead 
of adding another vertex that runs the MROutput. I'll follow up with this soon.
* POLocalRearrangeTez is added to each union source because 
ConcatenatedMergedKeyValuesInput expected ShuffledMergedInputs that requires 
sorting.


Diffs
-----

  src/org/apache/pig/backend/hadoop/executionengine/tez/POUnionTezLoad.java 
e496ca8 
  
src/org/apache/pig/backend/hadoop/executionengine/tez/POVertexGroupInputTez.java
 e69de29 
  src/org/apache/pig/backend/hadoop/executionengine/tez/TezCompiler.java 
245cade 
  src/org/apache/pig/backend/hadoop/executionengine/tez/TezDagBuilder.java 
bce8963 
  src/org/apache/pig/backend/hadoop/executionengine/tez/TezOperator.java 
b3aa020 
  src/org/apache/pig/backend/hadoop/executionengine/tez/TezPrinter.java f00946c 
  src/org/apache/pig/tools/pigstats/tez/TezStats.java feac11d 
  test/org/apache/pig/test/data/GoldenFiles/TEZC19.gld e69de29 
  test/org/apache/pig/tez/TestTezCompiler.java e71d838 

Diff: https://reviews.apache.org/r/19633/diff/


Testing
-------

ant test-tez passes except TestTezCompiler (known issue).
tez e2e tests all pass.


Thanks,

Cheolsoo Park

Reply via email to