[
https://issues.apache.org/jira/browse/HIVE-6710?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13941994#comment-13941994
]
Alan Gates commented on HIVE-6710:
----------------------------------
The issue is two fold:
# The code is not disciplined about ordering the access to the TXNS and
HIVE_LOCKS tables, this needs to be fixed
# Per the [mysql documentation |
https://dev.mysql.com/doc/refman/5.0/en/innodb-deadlocks.html], even when
tables are accessed in the right order deadlocks can be seen due to row locking
issues and thus the application should always be ready to re-start the
transaction. This is observed in the TxnManager.lock function when two clients
are obtaining locks at the same time.
> Deadlocks seen in transaction handler using mysql
> -------------------------------------------------
>
> Key: HIVE-6710
> URL: https://issues.apache.org/jira/browse/HIVE-6710
> Project: Hive
> Issue Type: Bug
> Components: Locking
> Affects Versions: 0.13.0
> Reporter: Alan Gates
> Assignee: Alan Gates
> Fix For: 0.13.0
>
>
> When multiple clients attempt to obtain locks a deadlock on the mysql
> database occasionally occurs.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
