[
https://issues.apache.org/jira/browse/IMPALA-13262?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17872477#comment-17872477
]
Fang-Yu Rao edited comment on IMPALA-13262 at 8/9/24 10:15 PM:
---------------------------------------------------------------
I started git bisecting from [IMPALA-9132: Explain statements should not cause
nullptr in
LogLineageRecord()|https://github.com/apache/impala/commit/f49f8d8a32] (which
is not affected by the bug) and it told us that the culprit is IMPALA-9979:
part 2.
In addition, setting '{*}ANALYTIC_RANK_PUSHDOWN_THRESHOLD{*}' to *0* could not
work around this issue.
{code:java}
fangyurao@fangyu:~/Impala_for_FE$ git bisect bad
b42c64993d46893488a667fb9c425548fdf964ab is the first bad commit
commit b42c64993d46893488a667fb9c425548fdf964ab
Author: Tim Armstrong <tarmstr...@cloudera.com>
Date: Tue Feb 2 14:02:12 2021 -0800
IMPALA-9979: part 2: partitioned top-n
{code}
was (Author: fangyurao):
I started git bisecting from [IMPALA-9132: Explain statements should not cause
nullptr in
LogLineageRecord()|https://github.com/apache/impala/commit/f49f8d8a32] (which
is not affected by the bug) and it told us that the culprit is IMPALA-9979:
part 2.
{code:java}
fangyurao@fangyu:~/Impala_for_FE$ git bisect bad
b42c64993d46893488a667fb9c425548fdf964ab is the first bad commit
commit b42c64993d46893488a667fb9c425548fdf964ab
Author: Tim Armstrong <tarmstr...@cloudera.com>
Date: Tue Feb 2 14:02:12 2021 -0800
IMPALA-9979: part 2: partitioned top-n
{code}
> Predicate pushdown causes incorrect results in join condition
> -------------------------------------------------------------
>
> Key: IMPALA-13262
> URL: https://issues.apache.org/jira/browse/IMPALA-13262
> Project: IMPALA
> Issue Type: Bug
> Reporter: Fang-Yu Rao
> Assignee: Fang-Yu Rao
> Priority: Major
> Labels: correctness
>
> We found that in some scenario Apache Impala
> (https://github.com/apache/impala/commit/c539874) could incorrectly push
> predicates to scan nodes, which in turn produces the wrong result. The
> following is a concrete example to reproduce the issue.
> {code:sql}
> create database impala_13262;
> use impala_13262;
> create table department ( dept_no integer, dept_rank integer, start_date
> timestamp,end_date timestamp);
> insert into department values(1,1,'2024-01-01','2024-01-02');
> insert into department values(1,2,'2024-01-02','2024-01-03');
> insert into department values(1,3,'2024-01-03','2024-01-03');
> create table employee (employee_no integer, depart_no integer);
> insert into employee values (1,1);
> // The following query should return 0 row. However Apache Impala produces
> one row.
> select * from employee t1
> inner join (
> select * from
> (
> select dept_no,dept_rank,start_date,end_date
> ,row_number() over(partition by dept_no order by dept_rank) rn
> from department
> ) t2
> where rn=1
> ) t2
> on t1.depart_no=t2.dept_no
> where t2.start_date=t2.end_date;
> set explain_level=2;
> // In the output of the EXPLAIN statement, we found that the predicate
> "start_data = end_date" was pushed
> // down to the scan node, which is wrong.
> | 01:SCAN HDFS [impala_13262.department, RANDOM]
> |
> | HDFS partitions=1/1 files=3 size=132B
> |
> | predicates: start_date = end_date
> |
> | stored statistics:
> |
> | table: rows=unavailable size=unavailable
> |
> | columns: unavailable
> |
> | extrapolated-rows=disabled max-scan-range-rows=unavailable
> |
> | mem-estimate=32.00MB mem-reservation=8.00KB thread-reservation=1
> |
> | tuple-ids=1 row-size=40B cardinality=1
> |
> | in pipelines: 01(GETNEXT)
> |
> +-------------------------------------------------------------------------------------------------------+
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
