[ 
https://issues.apache.org/jira/browse/HIVE-23368?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17099641#comment-17099641
 ] 

Zoltan Haindrich commented on HIVE-23368:
-----------------------------------------

[~jcamachorodriguez] I've just thinked about fixing this - but then I realized 
that it shouldn't even work "the best" with the rewrite:

if we have rewrites enabled:
{{count(distinct id)}} is rewritten to {{ds_hll_esitamte(ds_hll_sketch(id))}} 
which is from the MV point of view just like an int as {{count(distinct id)}} - 
so right now MV createrors should be aware if they want to create a sketch to 
be used by the rewrite.

If we want to automatically do a partial rewrite (save the binary instead of a 
plain int) - then we will most probably have a rule which changes the rowtype 
of a relnode; which might be problematic.

So right now I think it is safe to use MVs with datasketches rewrite.

At the time I was writing the qtest I've created the view storing a 
`ds_hll_sketch` of the input data; to enable rollups:
https://github.com/apache/hive/blob/5a119b9b556e1bfb0d7fdfcae2b5161b34d83f55/ql/src/test/queries/clientpositive/sketches_materialized_view_rollup2.q#L15


> MV rebuild should produce the same view as the one configured at creation time
> ------------------------------------------------------------------------------
>
>                 Key: HIVE-23368
>                 URL: https://issues.apache.org/jira/browse/HIVE-23368
>             Project: Hive
>          Issue Type: Sub-task
>            Reporter: Zoltan Haindrich
>            Assignee: Zoltan Haindrich
>            Priority: Major
>
> There might be some configrations which might affect the rel-tree of the 
> materialized views.
> In case rewrites to use datasketches for count(distinct) is enabled; the view 
> should store sketches instead of distinct values



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to