[
https://issues.apache.org/jira/browse/HIVE-2382?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Charles Chen updated HIVE-2382:
-------------------------------
Description:
When a GROUP BY is specified, a select operator is added before the GROUP BY in
SemanticAnalyzer.insertSelectAllPlanForGroupBy. Currently, the column
expression map for this is set to the column expression map for the parent
operator. This behavior is incorrect as, for example, the parent operator
could rearrange the order of the columns (_col0 => _col0, _col1 => _col2, _col2
=> _col1) and the new operator should not repeat this.
The predicate pushdown optimization uses the column expression map to track
which columns a filter expression refers to at different operators. This
results in a filter on incorrect columns.
Here is a simple case of this going wrong: Using
{noformat}
create table invites (id int, foo int, bar int);
{noformat}
executing the query
{noformat}
explain select * from (select foo, bar from (select bar, foo from invites c
union all select bar, foo from invites d) b) a group by bar, foo having bar=1;
{noformat}
results in
{noformat}
STAGE DEPENDENCIES:
Stage-1 is a root stage
Stage-0 is a root stage
STAGE PLANS:
Stage: Stage-1
Map Reduce
Alias -> Map Operator Tree:
a-subquery1:b-subquery1:c
TableScan
alias: c
Filter Operator
predicate:
expr: (foo = 1)
type: boolean
Select Operator
expressions:
expr: bar
type: int
expr: foo
type: int
outputColumnNames: _col0, _col1
Union
Select Operator
expressions:
expr: _col1
type: int
expr: _col0
type: int
outputColumnNames: _col0, _col1
Select Operator
expressions:
expr: _col0
type: int
expr: _col1
type: int
outputColumnNames: _col0, _col1
Group By Operator
bucketGroup: false
keys:
expr: _col1
type: int
expr: _col0
type: int
mode: hash
outputColumnNames: _col0, _col1
Reduce Output Operator
key expressions:
expr: _col0
type: int
expr: _col1
type: int
sort order: ++
Map-reduce partition columns:
expr: _col0
type: int
expr: _col1
type: int
tag: -1
a-subquery2:b-subquery2:d
TableScan
alias: d
Filter Operator
predicate:
expr: (foo = 1)
type: boolean
Select Operator
expressions:
expr: bar
type: int
expr: foo
type: int
outputColumnNames: _col0, _col1
Union
Select Operator
expressions:
expr: _col1
type: int
expr: _col0
type: int
outputColumnNames: _col0, _col1
Select Operator
expressions:
expr: _col0
type: int
expr: _col1
type: int
outputColumnNames: _col0, _col1
Group By Operator
bucketGroup: false
keys:
expr: _col1
type: int
expr: _col0
type: int
mode: hash
outputColumnNames: _col0, _col1
Reduce Output Operator
key expressions:
expr: _col0
type: int
expr: _col1
type: int
sort order: ++
Map-reduce partition columns:
expr: _col0
type: int
expr: _col1
type: int
tag: -1
Reduce Operator Tree:
Group By Operator
bucketGroup: false
keys:
expr: KEY._col0
type: int
expr: KEY._col1
type: int
mode: mergepartial
outputColumnNames: _col0, _col1
Select Operator
expressions:
expr: _col0
type: int
expr: _col1
type: int
outputColumnNames: _col0, _col1
File Output Operator
compressed: false
GlobalTableId: 0
table:
input format: org.apache.hadoop.mapred.TextInputFormat
output format:
org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
Stage: Stage-0
Fetch Operator
limit: -1
{noformat}
Note that the filter is now "foo = 1", while the correct behavior is to have
"bar = 1". If we remove the group by, the behavior is correct.
was:
When a GROUP BY is specified, a select operator is added before the GROUP BY in
SemanticAnalyzer.insertSelectAllPlanForGroupBy. Currently, the column
expression map for this is set to the column expression map for the parent
operator. This behavior is incorrect as, for example, the parent operator
could rearrange the order of the columns (_col0 => _col0, _col1 => _col2, _col2
=> _col1) and the new operator should not repeat this.
The predicate pushdown optimization uses the column expression map to track
which columns a filter expression refers to at different operators. This
results in a filter on incorrect columns.
Here is a simple case of this going wrong: Using
{noformat}
create table invites (id int, foo int, bar int)
{noformat}
executing the query
{noformat}
explain select * from (select foo, bar from (select bar, foo from invites c
union all select bar, foo from invites d) b) a group by bar, foo having bar=1;
{noformat}
results in
{noformat}
STAGE DEPENDENCIES:
Stage-1 is a root stage
Stage-0 is a root stage
STAGE PLANS:
Stage: Stage-1
Map Reduce
Alias -> Map Operator Tree:
a-subquery1:b-subquery1:c
TableScan
alias: c
Filter Operator
predicate:
expr: (foo = 1)
type: boolean
Select Operator
expressions:
expr: bar
type: int
expr: foo
type: int
outputColumnNames: _col0, _col1
Union
Select Operator
expressions:
expr: _col1
type: int
expr: _col0
type: int
outputColumnNames: _col0, _col1
Select Operator
expressions:
expr: _col0
type: int
expr: _col1
type: int
outputColumnNames: _col0, _col1
Group By Operator
bucketGroup: false
keys:
expr: _col1
type: int
expr: _col0
type: int
mode: hash
outputColumnNames: _col0, _col1
Reduce Output Operator
key expressions:
expr: _col0
type: int
expr: _col1
type: int
sort order: ++
Map-reduce partition columns:
expr: _col0
type: int
expr: _col1
type: int
tag: -1
a-subquery2:b-subquery2:d
TableScan
alias: d
Filter Operator
predicate:
expr: (foo = 1)
type: boolean
Select Operator
expressions:
expr: bar
type: int
expr: foo
type: int
outputColumnNames: _col0, _col1
Union
Select Operator
expressions:
expr: _col1
type: int
expr: _col0
type: int
outputColumnNames: _col0, _col1
Select Operator
expressions:
expr: _col0
type: int
expr: _col1
type: int
outputColumnNames: _col0, _col1
Group By Operator
bucketGroup: false
keys:
expr: _col1
type: int
expr: _col0
type: int
mode: hash
outputColumnNames: _col0, _col1
Reduce Output Operator
key expressions:
expr: _col0
type: int
expr: _col1
type: int
sort order: ++
Map-reduce partition columns:
expr: _col0
type: int
expr: _col1
type: int
tag: -1
Reduce Operator Tree:
Group By Operator
bucketGroup: false
keys:
expr: KEY._col0
type: int
expr: KEY._col1
type: int
mode: mergepartial
outputColumnNames: _col0, _col1
Select Operator
expressions:
expr: _col0
type: int
expr: _col1
type: int
outputColumnNames: _col0, _col1
File Output Operator
compressed: false
GlobalTableId: 0
table:
input format: org.apache.hadoop.mapred.TextInputFormat
output format:
org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
Stage: Stage-0
Fetch Operator
limit: -1
{noformat}
Note that the filter is now "foo = 1", while the correct behavior is to have
"bar = 1".
> Invalid predicate pushdown from incorrect column expression map for select
> operator generated by GROUP BY operation
> -------------------------------------------------------------------------------------------------------------------
>
> Key: HIVE-2382
> URL: https://issues.apache.org/jira/browse/HIVE-2382
> Project: Hive
> Issue Type: Bug
> Components: Query Processor
> Affects Versions: 0.6.0
> Reporter: Charles Chen
> Assignee: Charles Chen
> Priority: Critical
>
> When a GROUP BY is specified, a select operator is added before the GROUP BY
> in SemanticAnalyzer.insertSelectAllPlanForGroupBy. Currently, the column
> expression map for this is set to the column expression map for the parent
> operator. This behavior is incorrect as, for example, the parent operator
> could rearrange the order of the columns (_col0 => _col0, _col1 => _col2,
> _col2 => _col1) and the new operator should not repeat this.
> The predicate pushdown optimization uses the column expression map to track
> which columns a filter expression refers to at different operators. This
> results in a filter on incorrect columns.
> Here is a simple case of this going wrong: Using
> {noformat}
> create table invites (id int, foo int, bar int);
> {noformat}
> executing the query
> {noformat}
> explain select * from (select foo, bar from (select bar, foo from invites c
> union all select bar, foo from invites d) b) a group by bar, foo having bar=1;
> {noformat}
> results in
> {noformat}
> STAGE DEPENDENCIES:
> Stage-1 is a root stage
> Stage-0 is a root stage
> STAGE PLANS:
> Stage: Stage-1
> Map Reduce
> Alias -> Map Operator Tree:
> a-subquery1:b-subquery1:c
> TableScan
> alias: c
> Filter Operator
> predicate:
> expr: (foo = 1)
> type: boolean
> Select Operator
> expressions:
> expr: bar
> type: int
> expr: foo
> type: int
> outputColumnNames: _col0, _col1
> Union
> Select Operator
> expressions:
> expr: _col1
> type: int
> expr: _col0
> type: int
> outputColumnNames: _col0, _col1
> Select Operator
> expressions:
> expr: _col0
> type: int
> expr: _col1
> type: int
> outputColumnNames: _col0, _col1
> Group By Operator
> bucketGroup: false
> keys:
> expr: _col1
> type: int
> expr: _col0
> type: int
> mode: hash
> outputColumnNames: _col0, _col1
> Reduce Output Operator
> key expressions:
> expr: _col0
> type: int
> expr: _col1
> type: int
> sort order: ++
> Map-reduce partition columns:
> expr: _col0
> type: int
> expr: _col1
> type: int
> tag: -1
> a-subquery2:b-subquery2:d
> TableScan
> alias: d
> Filter Operator
> predicate:
> expr: (foo = 1)
> type: boolean
> Select Operator
> expressions:
> expr: bar
> type: int
> expr: foo
> type: int
> outputColumnNames: _col0, _col1
> Union
> Select Operator
> expressions:
> expr: _col1
> type: int
> expr: _col0
> type: int
> outputColumnNames: _col0, _col1
> Select Operator
> expressions:
> expr: _col0
> type: int
> expr: _col1
> type: int
> outputColumnNames: _col0, _col1
> Group By Operator
> bucketGroup: false
> keys:
> expr: _col1
> type: int
> expr: _col0
> type: int
> mode: hash
> outputColumnNames: _col0, _col1
> Reduce Output Operator
> key expressions:
> expr: _col0
> type: int
> expr: _col1
> type: int
> sort order: ++
> Map-reduce partition columns:
> expr: _col0
> type: int
> expr: _col1
> type: int
> tag: -1
> Reduce Operator Tree:
> Group By Operator
> bucketGroup: false
> keys:
> expr: KEY._col0
> type: int
> expr: KEY._col1
> type: int
> mode: mergepartial
> outputColumnNames: _col0, _col1
> Select Operator
> expressions:
> expr: _col0
> type: int
> expr: _col1
> type: int
> outputColumnNames: _col0, _col1
> File Output Operator
> compressed: false
> GlobalTableId: 0
> table:
> input format: org.apache.hadoop.mapred.TextInputFormat
> output format:
> org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat
> Stage: Stage-0
> Fetch Operator
> limit: -1
> {noformat}
> Note that the filter is now "foo = 1", while the correct behavior is to have
> "bar = 1". If we remove the group by, the behavior is correct.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
