[
https://issues.apache.org/jira/browse/HIVE-16997?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16096672#comment-16096672
]
Ashutosh Chauhan commented on HIVE-16997:
-----------------------------------------
IT would be nice to support following 3 paths of get_aggr_stats_for()
1. CachedStore disabled & useBitVectors: On this path we retrieve all rows from
RDBMS and do merging of bit vectors and aggregations of other stats in memory.
This is a new path being introduced in this patch.
2. CachedStore disable & no BitVectors: On this path we push aggr all the way
to RDBMS. This was the path before this patch.
3. CachedStore enabled : Merging of stats is already there since it uses hbase
merging logic. However, it is missing initial load of stats data since that
happens via RawStore::getColStatsForTablePartitions() which calls into
DirectSql::getColStatsForTablePartitions() which needs to be updated to
retrieve bitvectors as well.
And then there is extrapolation logic. I think we shall refactor extrapolation
so that its usable after stats merging in memory or after pushed aggregation to
RDBMS.
> Extend object store to store bit vectors
> ----------------------------------------
>
> Key: HIVE-16997
> URL: https://issues.apache.org/jira/browse/HIVE-16997
> Project: Hive
> Issue Type: Sub-task
> Reporter: Pengcheng Xiong
> Assignee: Pengcheng Xiong
> Attachments: HIVE-16997.01.patch, HIVE-16997.02.patch
>
>
> This patch includes: (1) a new serde for FMSketch (2) change of schema for
> derby and mysql (3) support for date type (4) refactoring the extrapolation
> and merge code
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
