Paul Rogers created DRILL-7080:
----------------------------------
Summary: Inconsistent behavior with wildcard and partition columns
Key: DRILL-7080
URL: https://issues.apache.org/jira/browse/DRILL-7080
Project: Apache Drill
Issue Type: Bug
Affects Versions: 1.15.0
Reporter: Paul Rogers
Drill supports queries of the form:
{code:sql}
SELECT *, dir0 FROM `myTable`
{code}
Where `myTable` is, say, a set of CSV files with columns "a", "b" and "c". As
shown in the (soon to be submitted) {{TestCsvWithHeaders}} test, behavior of
partition columns is wildly inconsistent and nearly unusable. This ticket focus
on one specific issue: the query above results in a schema like (dir0, a, b, c,
dir00). That is:
* The wildcard generates "dir0", "dir1" columns.
* The Project operator inserts a second column, "dir00" as type Nullable Int.
This behavior is surprising as the following query produces the expected result:
{code:sql}
SELECT *, filename from `myTable`
{code}
That is, the above produces a schema of the form (a, b, c, filename) with
"filename" of the expected type: VARCHAR.
This appears to be a bug somewhere in the project operator and/or the planner,
but I've not tracked down the root cause.
The workaround is to either:
1. Not include the "dir0" column explicitly with the wildcard, or
2. Don't use the wildcard: list columns explicitly, including the partition
columns.
Given how late in the game that this bug is filed, I would guess that few
people actually use this feature.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
