[
https://issues.apache.org/jira/browse/DRILL-1847?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jason Altekruse updated DRILL-1847:
-----------------------------------
Description: Union all currently relies on columns to be named the same to
merge them, this does not follow the behavior of traditional databases. This is
a bit complicated with Drill, as we do not always know type information up
front, so validating these queries will have to be less strict. We need to
either fix this at planning to insert renaming based on the column orderings in
the query or by changing the behavior of the operator to use the column orders
that come in. This does not seem to be the best behavior as the engine
canonicalizes the column order in some cases, so the order is not always
preserved in the middle of processing, it is reordered at the end to provide
the correct order to users. (was: Union all currently relies on columns to be
named the same to merge them, this does not follow the behavior of traditional
databases. This is a bit complicated with Drill, as we do not always know type
information up front, so validating these queries will have to less strict. We
need to either fix this at planning to insert renaming based on the column
orderings in the query or by changing the behavior of the operator to use the
column orders that come in. This does not seem to be the best behavior as the
engine canonicalizes the column order in some cases, so the order is not always
preserved in the middle of processing, it is reordered at the end to provide
the correct order to users.)
> Union All does not match corresponding columns by index to union columns
> ------------------------------------------------------------------------
>
> Key: DRILL-1847
> URL: https://issues.apache.org/jira/browse/DRILL-1847
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Jason Altekruse
>
> Union all currently relies on columns to be named the same to merge them,
> this does not follow the behavior of traditional databases. This is a bit
> complicated with Drill, as we do not always know type information up front,
> so validating these queries will have to be less strict. We need to either
> fix this at planning to insert renaming based on the column orderings in the
> query or by changing the behavior of the operator to use the column orders
> that come in. This does not seem to be the best behavior as the engine
> canonicalizes the column order in some cases, so the order is not always
> preserved in the middle of processing, it is reordered at the end to provide
> the correct order to users.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
