[
https://issues.apache.org/jira/browse/HIVE-9422?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15172907#comment-15172907
]
Sergey Shelukhin commented on HIVE-9422:
----------------------------------------
Sorry for the delay.
1) In general, this approach makes sense. The outer loop (for (int i = 0; i <
maxBatchesRG; i++)) logic might need to change since the number of batches
might change. Probably good to run some tests on this with lots of rows
filtered out.
2) sargApp.pickRow(cvb, sarged_cvb); - the implementation for this is not
included. Note that the next optimization would be to filter based only on
relevant vectors, and then filling the others; that might be relevant for
pickRow implementation.
3) assert sarged_cvb.size >= cvb.size; - wouldn't this be false if some rows
are filtered out.
> LLAP: row-level vectorized SARGs
> --------------------------------
>
> Key: HIVE-9422
> URL: https://issues.apache.org/jira/browse/HIVE-9422
> Project: Hive
> Issue Type: Sub-task
> Components: llap
> Reporter: Sergey Shelukhin
> Attachments: HIVE-9422.WIP1.patch
>
>
> When VRBs are built from encoded data, sargs can be applied on low level to
> reduce the number of rows to process.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
