[
https://issues.apache.org/jira/browse/CRUNCH-286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13805713#comment-13805713
]
Gabriel Reid commented on CRUNCH-286:
-------------------------------------
I don't think it's really necessary to make the in-memory impl replicate the MR
impl -- it's technically possible that an MR map-phase combiner wouldn't get
any multiple values to combine, so I think it's safe to just discard the
map-phase CombineFn in the mem-pipeline impl.
I'm thinking a mempipline doesn't execute a CombineFn in something like a
map-phase anyway, so it probably would make much of a difference there. Or
would it?
> ability to specify a different function for combiner & reducer
> --------------------------------------------------------------
>
> Key: CRUNCH-286
> URL: https://issues.apache.org/jira/browse/CRUNCH-286
> Project: Crunch
> Issue Type: New Feature
> Components: Core
> Reporter: Stefan De Smit
> Assignee: Josh Wills
> Priority: Minor
> Attachments:
> 0001-add-combineValues-method-with-2-function-arguments.patch, 0002-.patch,
> CRUNCH-286-prototype.patch
>
>
> Extend PGroupedTable with an extra combineValues function that accepts 2
> functions: 1 for combiner phase, 1 for reducer phase.
> This way, different algorithm can be applied.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
