[jira] [Comment Edited] (HIVE-23712) metadata-only queries return incorrect results with empty partition

["László Bodor (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-23712?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17138436#comment-17138436 ] 

László Bodor edited comment on HIVE-23712 at 6/17/20, 1:35 PM:
---------------------------------------------------------------

the root cause is that in [MetadataOnlyOptimizer|https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/optimizer/physical/MetadataOnlyOptimizer.java#L124] the TS operator of test1 table is considered to be subject of metadata-only optimization and later [NullScanTaskDispatcher|https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/optimizer/physical/NullScanTaskDispatcher.java#L106] find a non-empty folder for this partition because of ACID operations, with files:
{code}
hdfs://localhost:58447/build/ql/test/data/warehouse/test1/val2=bar/delete_delta_0000003_0000003_0000
hdfs://localhost:58447/build/ql/test/data/warehouse/test1/val2=bar/delta_0000002_0000002_0000
{code}

not sure about the perfect solution at the moment, but maybe the following scenario should be excluded somehow from metadata-only optimization:
1. there is a partitioned table:
create table test1 (id int, val string) partitioned by (val2 string) STORED AS ORC TBLPROPERTIES ('transactional'='true');
2. in a distinct query, only the partitioned column is selected:
{code}
select distinct val2, current_timestamp, 'metadata true' as query from test1;
{code}
in this case tsOp.getNeededColumnIDs() is empty (partition column is not present in needed columns)



was (Author: abstractdog):
the root cause is that in [MetadataOnlyOptimizer|https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/optimizer/physical/MetadataOnlyOptimizer.java#L124] the TS operator of test1 table is considered to be subject of metadata-only optimization and later [NullScanTaskDispatcher|https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/optimizer/physical/NullScanTaskDispatcher.java#L106] find a non-empty folder for this partition because of ACID operations, with files:
{code}
hdfs://localhost:58447/build/ql/test/data/warehouse/test1/val2=bar/delete_delta_0000003_0000003_0000
hdfs://localhost:58447/build/ql/test/data/warehouse/test1/val2=bar/delta_0000002_0000002_0000
{code}

not sure about the perfect solution at the moment, but maybe the following scenario should be excluded somehow from metadata-only optimization:
1. there is a partitioned table:
create table test1 (id int, val string) partitioned by (val2 string) STORED AS ORC TBLPROPERTIES ('transactional'='true');
2. in a distinct query, only the partitioned column is selected:
select distinct val2, current_timestamp, 'metadata true' as query from test1;
{code}
in this case tsOp.getNeededColumnIDs() is empty (partition column is not present in needed columns)
{code}


> metadata-only queries return incorrect results with empty partition
> -------------------------------------------------------------------
>
>                 Key: HIVE-23712
>                 URL: https://issues.apache.org/jira/browse/HIVE-23712
>             Project: Hive
>          Issue Type: Bug
>            Reporter: László Bodor
>            Assignee: László Bodor
>            Priority: Major
>
> Similarly to HIVE-15397, queries can return incorrect results for metadata-only queries, here is a repro scenario which affects master:
> {code}
> set hive.support.concurrency=true;
> set hive.exec.dynamic.partition.mode=nonstrict;
> set hive.txn.manager=org.apache.hadoop.hive.ql.lockmgr.DbTxnManager;
> set hive.optimize.metadataonly=true;
> create table test1 (id int, val string) partitioned by (val2 string) STORED AS ORC TBLPROPERTIES ('transactional'='true');
> describe formatted test1;
> alter table test1 add partition (val2='foo');
> alter table test1 add partition (val2='bar');
> insert into test1 partition (val2='foo') values (1, 'abc');
> select distinct val2, current_timestamp from test1;
> insert into test1 partition (val2='bar') values (1, 'def');
> delete from test1 where val2 = 'bar';
> select '--> hive.optimize.metadataonly=true';
> select distinct val2, current_timestamp from test1;
> set hive.optimize.metadataonly=false;
> select '--> hive.optimize.metadataonly=false';
> select distinct val2, current_timestamp from test1;
> select current_timestamp, * from test1;
> {code}
> in this case 2 rows returned instead of 1 after a delete with metadata only optimization:
> https://github.com/abstractdog/hive/commit/a7f03513564d01f7c3ba4aa61c4c6537100b4d3f#diff-cb23043000831f41fe7041cb38f82224R114-R128



--
This message was sent by Atlassian Jira
(v8.3.4#803005)