[
https://issues.apache.org/jira/browse/TEZ-1080?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14046279#comment-14046279
]
Bikas Saha commented on TEZ-1080:
---------------------------------
I like that if we use the OrderedPartitionedEdgeConfiguration then it hides the
implementation of the actual input and output. Does this also imply Key-Value?
If yes, then we could add that to the name. After that, does processor need to
know the input/output impls or can it simply get away by doing (KVReader)
input.getReader()?
Please make sure that non-public classes are marked as such. A lot of classes
have been added and exposing them would not be good.
The default for any partitioned edge if fine to be of type SCATTER_GATHER.
Though for UnorderedUnpartitionedEdgeConfiguration, broadcast and 1-1 are both
equally applicable. There may be areas where not having a default might make
sense or have 2 methods for defaultBroadcast and defaultOneToOne.
Overall this is a step forward. We may have to iron out a bit once
Hive/Pig/others actually use this. I would really like to remove the older
method of configuration so that we can get rid of that baggage.
> Configuration for non MR based Inputs/Outputs
> ---------------------------------------------
>
> Key: TEZ-1080
> URL: https://issues.apache.org/jira/browse/TEZ-1080
> Project: Apache Tez
> Issue Type: Sub-task
> Reporter: Siddharth Seth
> Assignee: Siddharth Seth
> Attachments: TEZ-1080.4.txt, TEZ-1080.wip.1.txt, TEZ-1080.wip.2.txt,
> TEZ-1080.wip.3.txt
>
>
> De-link configuration from MRHelpers (except for the YARNRunner case), and
> allow for these to be configured easily - exposing necessary setters /
> getters without having to rely on config keys.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
