hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Zoltan Haindrich (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HIVE-19889) Wrong results due to PPD of non deterministic functions with CBO
Date Thu, 14 Jun 2018 08:55:00 GMT

     [ https://issues.apache.org/jira/browse/HIVE-19889?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Zoltan Haindrich updated HIVE-19889:
------------------------------------
    Description: 
The following query can give wrong results when CBO is on:
{code}
select * from (
select part1,randum123
from (SELECT *, cast(rand() as double) AS randum123 FROM testA where part1='CA' and part2
= 'ABC') a
where randum123 <= 0.5) s where s.randum123 > 0.25 limit 20;

The plan of the query is as follows:
STAGE PLANS:
  Stage: Stage-1
    Map Reduce
      Map Operator Tree:
          TableScan
            alias: testa
            Statistics: Num rows: 2 Data size: 4580 Basic stats: COMPLETE Column stats: NONE
            Filter Operator
              predicate: ((rand() <= 0.5D) and (rand() > 0.25D)) (type: boolean)
              Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
              Select Operator
                expressions: 'CA' (type: string), rand() (type: double)
                outputColumnNames: _col0, _col1
                Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
                Limit
                  Number of rows: 20
                  Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
                  File Output Operator
                    compressed: false
                    Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
                    table:
                        input format: org.apache.hadoop.mapred.SequenceFileInputFormat
                        output format: org.apache.hadoop.hive.ql.io.HiveSequenceFileOutputFormat
                        serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe

  Stage: Stage-0
    Fetch Operator
      limit: 20
      Processor Tree:
        ListSink
{code}

The relevant part in the plan is the filter:

{code}
            Filter Operator
              predicate: ((rand() <= 0.5D) and (rand() > 0.25D)) (type: boolean)
{code}

The predicates s.randum123 > 0.25 and s.randum123 > 0.25 were pushed down.  And randum123
was resolved to rand().  This is bad because it will result in invocation of rand() two times
and rand() UDF is non-deterministic.  Both the rand calls can generate values that can satisfy
the predicates independently, but not together, whereas the original intention of the query
is to give results when rand falls between 0.25 and 0.5.

A sample result:

{code}
CA	0.9191984370369802
CA	0.397933021566812
{code}

where the condition was not satisfied.

  was:
The following query can give wrong results when CBO is on:
select * from (
select part1,randum123
from (SELECT *, cast(rand() as double) AS randum123 FROM testA where part1='CA' and part2
= 'ABC') a
where randum123 <= 0.5) s where s.randum123 > 0.25 limit 20;

The plan of the query is as follows:
STAGE PLANS:
  Stage: Stage-1
    Map Reduce
      Map Operator Tree:
          TableScan
            alias: testa
            Statistics: Num rows: 2 Data size: 4580 Basic stats: COMPLETE Column stats: NONE
            Filter Operator
              predicate: ((rand() <= 0.5D) and (rand() > 0.25D)) (type: boolean)
              Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
              Select Operator
                expressions: 'CA' (type: string), rand() (type: double)
                outputColumnNames: _col0, _col1
                Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
                Limit
                  Number of rows: 20
                  Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
                  File Output Operator
                    compressed: false
                    Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
                    table:
                        input format: org.apache.hadoop.mapred.SequenceFileInputFormat
                        output format: org.apache.hadoop.hive.ql.io.HiveSequenceFileOutputFormat
                        serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe

  Stage: Stage-0
    Fetch Operator
      limit: 20
      Processor Tree:
        ListSink

The relevant part in the plan is the filter:
            Filter Operator
              predicate: ((rand() <= 0.5D) and (rand() > 0.25D)) (type: boolean)

The predicates s.randum123 > 0.25 and s.randum123 > 0.25 were pushed down.  And randum123
was resolved to rand().  This is bad because it will result in invocation of rand() two times
and rand() UDF is non-deterministic.  Both the rand calls can generate values that can satisfy
the predicates independently, but not together, whereas the original intention of the query
is to give results when rand falls between 0.25 and 0.5.

A sample result:
CA	0.9191984370369802
CA	0.397933021566812
where the condition was not satisfied.


> Wrong results due to PPD of non deterministic functions with CBO
> ----------------------------------------------------------------
>
>                 Key: HIVE-19889
>                 URL: https://issues.apache.org/jira/browse/HIVE-19889
>             Project: Hive
>          Issue Type: Bug
>            Reporter: Janaki Lahorani
>            Assignee: Janaki Lahorani
>            Priority: Major
>             Fix For: 4.0.0
>
>         Attachments: HIVE-19889.1.patch
>
>
> The following query can give wrong results when CBO is on:
> {code}
> select * from (
> select part1,randum123
> from (SELECT *, cast(rand() as double) AS randum123 FROM testA where part1='CA' and part2
= 'ABC') a
> where randum123 <= 0.5) s where s.randum123 > 0.25 limit 20;
> The plan of the query is as follows:
> STAGE PLANS:
>   Stage: Stage-1
>     Map Reduce
>       Map Operator Tree:
>           TableScan
>             alias: testa
>             Statistics: Num rows: 2 Data size: 4580 Basic stats: COMPLETE Column stats:
NONE
>             Filter Operator
>               predicate: ((rand() <= 0.5D) and (rand() > 0.25D)) (type: boolean)
>               Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column stats:
NONE
>               Select Operator
>                 expressions: 'CA' (type: string), rand() (type: double)
>                 outputColumnNames: _col0, _col1
>                 Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column
stats: NONE
>                 Limit
>                   Number of rows: 20
>                   Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column
stats: NONE
>                   File Output Operator
>                     compressed: false
>                     Statistics: Num rows: 1 Data size: 2290 Basic stats: COMPLETE Column
stats: NONE
>                     table:
>                         input format: org.apache.hadoop.mapred.SequenceFileInputFormat
>                         output format: org.apache.hadoop.hive.ql.io.HiveSequenceFileOutputFormat
>                         serde: org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe
>   Stage: Stage-0
>     Fetch Operator
>       limit: 20
>       Processor Tree:
>         ListSink
> {code}
> The relevant part in the plan is the filter:
> {code}
>             Filter Operator
>               predicate: ((rand() <= 0.5D) and (rand() > 0.25D)) (type: boolean)
> {code}
> The predicates s.randum123 > 0.25 and s.randum123 > 0.25 were pushed down.  And
randum123 was resolved to rand().  This is bad because it will result in invocation of rand()
two times and rand() UDF is non-deterministic.  Both the rand calls can generate values that
can satisfy the predicates independently, but not together, whereas the original intention
of the query is to give results when rand falls between 0.25 and 0.5.
> A sample result:
> {code}
> CA	0.9191984370369802
> CA	0.397933021566812
> {code}
> where the condition was not satisfied.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message