[
https://issues.apache.org/jira/browse/HIVE-17645?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16211886#comment-16211886
]
Sergey Shelukhin commented on HIVE-17645:
-----------------------------------------
Spark+ACID only uses the hacky mode for select queries so it should be ok as
long as we don't get it from the session for selects.
However, a larger concern I have is this... how does it work at all if a
different TxnManager has non shared state with the main one? They'd be able to
take locks separately in parallel for the same things.
And if they don't have non-shared state (rely on the same metastore DB, ZK/DB
lock paths, etc) then what's the problem with getting a different txn manager?
> MM tables patch conflicts with HIVE-17482 (Spark/Acid integration)
> ------------------------------------------------------------------
>
> Key: HIVE-17645
> URL: https://issues.apache.org/jira/browse/HIVE-17645
> Project: Hive
> Issue Type: Sub-task
> Components: Transactions
> Affects Versions: 3.0.0
> Reporter: Eugene Koifman
> Labels: mm-gap-2
>
> MM code introduces
> {noformat}
> HiveTxnManager txnManager = SessionState.get().getTxnMgr()
> {noformat}
> in a number of places (e.g _DDLTask.generateAddMmTasks(Table tbl)_).
> HIVE-17482 adds a mode where a TransactionManager not associated with the
> session should be used. This will need to be addressed.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
