[
https://issues.apache.org/jira/browse/HIVE-19617?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16484136#comment-16484136
]
Jesus Camacho Rodriguez commented on HIVE-19617:
------------------------------------------------
[~kgyrtkirk], thanks for the feedback. I do not think we ever had a mechanism
to reset Hive to its initial state between tests when we chain their execution:
{code}
mvn test -Dtest=TestCliDriver -Dqfile=test1.q,test2.q
-Dtest.output.overwrite=true
{code}
Devs have been seeing this kind of interaction among tests before, e.g., I am
sure you have seen that many q file tests start with DROP TABLE {{t}} if {{t}}
is going to be created just afterwards, but {{t}} should not exist if the test
is executed in isolation.
Making the names unique is certainly a workaround, and this patch seems to help
to get rid of the flakiness. I think the correct way to go is what you propose,
but that needs some work to identify those changes that have happened during
Hive execution; I do not think you want to tear down and initialize the
metastore for each individual test execution since that will impact test
execution time.
> Rename test tables to avoid collisions during execution in batches
> ------------------------------------------------------------------
>
> Key: HIVE-19617
> URL: https://issues.apache.org/jira/browse/HIVE-19617
> Project: Hive
> Issue Type: Test
> Components: Testing Infrastructure
> Affects Versions: 3.1.0
> Reporter: Jesus Camacho Rodriguez
> Assignee: Jesus Camacho Rodriguez
> Priority: Critical
> Fix For: 3.1.0
>
> Attachments: HIVE-19617.01.patch, HIVE-19617.02.patch,
> HIVE-19617.03.patch, HIVE-19617.patch
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
