[jira] [Comment Edited] (CALCITE-4599) Support push-down of "date histogram aggregation" in elastic search adapter

["Jacky Yin (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CALCITE-4599?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17370444#comment-17370444 ] 

Jacky Yin edited comment on CALCITE-4599 at 6/29/21, 2:53 AM:
--------------------------------------------------------------

Hello [~julianhyde],

The TVF 'tumble' can be used without aggregation. However, the date_histogram function is only supported with aggregation in elastic search. So it seems not proper to simply push down the TVF 'tumble'. How about the legacy window grouping function 'tumble' like below? 
{code:java}
select tumble_start(ts, interval '1' minute) as window_s, count(*) as cc from table1 group by tumble(ts, interval '1' minute){code}
I know it will be deprecated in the future. So is it still worthwhile to implement it? 


was (Author: jackyyin):
Hello [~julianhyde],

The TVF 'tumble' can be used without aggregation. However, the date_histogram function is only supported with aggregation in elastic search. So it seems not proper to push down the TVF 'tumble'. How about the legacy window grouping function 'tumble' like below? 
{code:java}
select tumble_start(ts, interval '1' minute) as window_s, count(*) as cc from table1 group by tumble(ts, interval '1' minute){code}

> Support push-down of "date histogram aggregation" in elastic search adapter
> ---------------------------------------------------------------------------
>
>                 Key: CALCITE-4599
>                 URL: https://issues.apache.org/jira/browse/CALCITE-4599
>             Project: Calcite
>          Issue Type: Improvement
>          Components: elasticsearch-adapter
>            Reporter: Jacky Yin
>            Priority: Critical
>
> "date histogram aggregation" is one of the most popular analysis function of elastic search. It is not pushed down to elastic search in current calcite es adapter. Is there any plan to support it? Given the below example, if it can be pushed down to es, the query should be very efficient. 
> {code:java}
> select count as cc from t group by date_histogram(`@timestamp`, interval '5' minute){code}
> Another question is currently there seems no proper sql function/keyword for date_histogram.  One possible option is TUMBLE function. Is it right? 



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