[
https://issues.apache.org/jira/browse/BEAM-10216?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rui Wang updated BEAM-10216:
----------------------------
Description:
Current TVF Session has the following limitation:
Use a "dummy" key to go through a GBK to compute window_start and window_end.
This introduces an extra shuffle so it costs more.
One optimizations we can do is always merge TableFunctionScanRel with
AggregationRel (and also all Join Rel), thus we can use only one GBK to compute
both SESSION window_start and window_end, and aggregations(joins).
was:
Current TVF Session has the following limitation:
Use a "dummy" key to go through a GBK to compute window_start and window_end.
This introduces an extra shuffle so it costs more.
One optimizations we can do is always merge TableFunctionScanRel with
AggregationRel (and also all Join Rel), thus we can use only one GBK to compute
both SESSION window_start and window_end, and aggregations.
> SESSION TVF Optimizations
> -------------------------
>
> Key: BEAM-10216
> URL: https://issues.apache.org/jira/browse/BEAM-10216
> Project: Beam
> Issue Type: Improvement
> Components: dsl-sql-zetasql
> Reporter: Rui Wang
> Priority: P2
>
> Current TVF Session has the following limitation:
> Use a "dummy" key to go through a GBK to compute window_start and window_end.
> This introduces an extra shuffle so it costs more.
> One optimizations we can do is always merge TableFunctionScanRel with
> AggregationRel (and also all Join Rel), thus we can use only one GBK to
> compute both SESSION window_start and window_end, and aggregations(joins).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
