[
https://issues.apache.org/jira/browse/HAWQ-68?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14964426#comment-14964426
]
ASF GitHub Bot commented on HAWQ-68:
------------------------------------
Github user jianlirong commented on the pull request:
https://github.com/apache/incubator-hawq/pull/29#issuecomment-149413822
As the JIRA description suggested, the strategy of this fix is to let the
partition child table with the largest data size decide whether all other
partition child tables should be transformed from hash to random.
According to this definition, this fix is OK to me. +1.
> Wrong answer when converting hash to random partition table
> -----------------------------------------------------------
>
> Key: HAWQ-68
> URL: https://issues.apache.org/jira/browse/HAWQ-68
> Project: Apache HAWQ
> Issue Type: Bug
> Components: Core
> Reporter: Hubert Zhang
> Assignee: Hubert Zhang
>
> for hash partition table, if datalocality is below a threshold, it will be
> treated as random. For now, some partition's datalocality are high, some are
> not, which leads to some partitions are converted to random, some are not.
> But optimizer must treat the whole partition table with the same distribution
> type. So we determine the distribution type of partition table by its first
> partition.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
