[HIVE-18390] IndexOutOfBoundsException when query a partitioned view in ColumnPruner

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1.1
• Fix Version/s: 3.0.0
• Component/s: Query Planning, Views
• Assignee: Hengyu Dai
• Reporter: Hengyu Dai
• Due: 06/Jan/18
• Created: 06/Jan/18 12:13
• Updated: 22/May/18 23:14
• Resolved: 09/Jan/18 18:14

Description

IndexOutOfBoundsException is encountered when query a partitioned view.
in Column Prunning, each SEL operator collects the accessed column in current SEL operator,
When ColumnPrunerSelectProc getting a view's columns accessed, it will first get the index of output column names in the view, then call Table.getCols().get(index).getName() to finally get the
name of output column, but Table.getCols() will not return all columns (partitioned column is
lacked), so if partitioned columns is queried, an IndexOutOfBoundsException will throw.

REPRODUCE:

create table foo
(
`a` string
) partitioned by (`b` string)
;

create view bar partitioned on (b) as
select a,b from foo;

select * from bar;     --IndexOutOfBoundsException

OPERATORE TREE:

TS[0]
   |
SEL[1]
   |
SEL[2]
   |
FS[3]

SEL[1] collects accessed column(contains partitioned column b), b's internal column name is '_col1', the corresponding column index is 1, but actually bar's getCols() returned a list of length 1: ['a'], so tab.getCols().get(1) throw tab.getCols().get(index)

HOW TO FIX:
instead of call view's getCols() method, we should get all columns including partitioned columns