Improve UDAF support for constant OI and better initialization
--------------------------------------------------------------
Key: HIVE-2541
URL: https://issues.apache.org/jira/browse/HIVE-2541
Project: Hive
Issue Type: Improvement
Components: UDF
Reporter: Igor Kabiljo
Currently, GenericUDAFs have only basic support for constant OIs - you can have
constant OI as input, and check it in getEvaluator and in init() for mode where
input is not partial, and that is all.
For it to be fully useful, it would be great to have (not sure if this is best
as 3 separate tickets, or one):
- if output of terminate and terminatePartial is constantOI (or partially
constant OI, ie struct with some constant and some non-constant fields), input
of merge and next function should receive it as constant OI.
- add initUDAFArguments(GenericUDAFParameterInfo info) to
GenericUDAFEvaluator, that will be always called before init(Mode m,
ObjectInspector[] argOIs). That would allow you to access constant OI
information in all modes (currently in PARTIAL2 and FINAL modes it is not
possible). GenericUDAFEvaluator should have default empty implementation, so
that only classes that need that info can access it.
- currently, when GenericUDAFEvaluator is returned by getEvaluator, all
initialization done there (in constructor or elsewhere), is removed by
serializing/deserializing it. (ie new instance of evaluator created by calling
no args constructor is used). It would be great if fields would not be lost.
This can be either achieved by fixing serialization/deserialization to save all
fields as well (if that is possible by XMLEncoder), or by
serializing/deserializing AbstractGenericUDAFResolver instead of Evaluator, and
then calling getEvaluator whenever Evaluator is needed.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
