[
https://issues.apache.org/jira/browse/HIVE-8651?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14190509#comment-14190509
]
Ashutosh Chauhan commented on HIVE-8651:
----------------------------------------
Thinking more about this. I think this is not an issue. CBO reordered the join
and Hive still correctly inferred bucketing / sorting properties. Since, these
properties are recorded per partition and SMBJ & BMJ pay attention to per
partition properties, bucketing of tables and subsequent joins should still
work fine. More info on HIVE-933
cc: [~vikram.dixit] If different partitions are bucketed / sorted on different
keys what happens to SMBJ / BMJ? As per my understanding, those optimized joins
make checks for all partitions and kicks in only if all partitions are bucketed
/ sorted on same key. Can you confirm?
If thats true, than I think this is a non-issue. Although, we should update
comment on that test (which tripped me) to reflect new reality.
> CBO: sort column changed in infer_bucket_sort test
> --------------------------------------------------
>
> Key: HIVE-8651
> URL: https://issues.apache.org/jira/browse/HIVE-8651
> Project: Hive
> Issue Type: Sub-task
> Components: CBO
> Reporter: Sergey Shelukhin
> Assignee: Jesús Camacho Rodríguez
> Fix For: 0.15.0
>
>
> {noformat}
> -Bucket Columns: [value]
> -Sort Columns: [Order(col:value, order:1)]
> +Bucket Columns: [key]
> +Sort Columns: [Order(col:key, order:1)]
> {noformat}
> That is in the case for "Test join on three tables on different keys, should
> be bucketed and sorted by latter key"; the sort changed to the first key
> instead
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
