[
https://issues.apache.org/jira/browse/DRILL-335?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13921911#comment-13921911
]
Aman Sinha commented on DRILL-335:
----------------------------------
An updated patch which contains the cumulative changes was posted to
reviewboard. This second patch contains following enhancements/fixes:
1. It includes support for the code generation of the probe side of a hash join
(previously, we only needed the 'build' side for the hash aggregate)
2. Fix for a index out of bound error that I encountered while doing testing -
this was related to using the wrong TypedFieldId when accessing the hash
table's container. This is resolved by doing the creation of htContainer as
part of ChainedHashTable and making a clone of it in HashTableTemplate.
3. Address a review comment: Allocate the links, hashValues and startIndices
using our vector allocators (off-heap) instead of on the heap.
> Implement hash aggregation operator for grouped aggregation
> -----------------------------------------------------------
>
> Key: DRILL-335
> URL: https://issues.apache.org/jira/browse/DRILL-335
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Aman Sinha
> Assignee: Aman Sinha
> Attachments: Drill-335-1.patch, Drill-335-2.patch, DrillAggrs.pdf
>
>
> Placeholder for doing design and implementation of hash based aggregation in
> Drill. Currently, it supports sort based aggregation. Hash aggregation
> involves building a hash table on the grouping columns and maintaining
> workspace for each of the aggregate functions in the query. This issue will
> focus on non-distinct aggregation for now (i.e aggregate function does not
> contain DISTINCT). A separate issue will be created to track distinct
> aggregation.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
