[
https://issues.apache.org/jira/browse/HIVE-13286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15196044#comment-15196044
]
Vikram Dixit K commented on HIVE-13286:
---------------------------------------
[~aihuaxu] Consider the following scenario: In Tez/Spark, if we end up caching
the small table based on the hive query id. If say the user set the hive query
id for 1 query and does not reset it for the subsequent query, we will end up
picking the previously cached hash table for the join resulting in incorrect
results right? Creating a new conf object would only work if we reset the query
id after the query completes. If we allow it to exist in the configuration
object after a query has completed running, it will result in incorrect results
or some weird behavior.
Consider hs2 or cli session, if a user in a session assigns a query id and
doesn't reset it, it can result in incorrect results. You are expecting a user
to set a query id each time after setting it once? I don't think that is great
behavior.
> Query ID is being reused across queries
> ---------------------------------------
>
> Key: HIVE-13286
> URL: https://issues.apache.org/jira/browse/HIVE-13286
> Project: Hive
> Issue Type: Bug
> Components: Parser
> Affects Versions: 2.0.0
> Reporter: Vikram Dixit K
> Assignee: Aihua Xu
> Priority: Critical
>
> [~aihuaxu] I see this commit made via HIVE-11488. I see that query id is
> being reused across queries. This defeats the purpose of a query id. I am not
> sure what the purpose of the change in that jira is but it breaks the
> assumption about a query id being unique for each query. Please take a look
> into this at the earliest.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
