[
https://issues.apache.org/jira/browse/HIVE-11848?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15269526#comment-15269526
]
Eugene Koifman commented on HIVE-11848:
---------------------------------------
reran all failed tests with age=1 All passed
> tables in subqueries don't get locked
> -------------------------------------
>
> Key: HIVE-11848
> URL: https://issues.apache.org/jira/browse/HIVE-11848
> Project: Hive
> Issue Type: Bug
> Components: Locking, Transactions
> Affects Versions: 1.0.0
> Reporter: Eugene Koifman
> Assignee: Eugene Koifman
> Priority: Critical
> Attachments: HIVE-11848.patch
>
>
> Consider
> {noformat}
> update acidTbl set b=19 where acidTbl.b in(select I.b from nonAcidOrcTbl I
> where I.a = 3)
> {noformat}
> noAcidOrcTbl doesn't get locked at all. (SHARED_WRITE is taken on acidTbl).
> Same for __delete__ with subquery
> This is is because the ReadEntity for nonAcidOrcTbl is skipped by
> {noformat}
> for (ReadEntity input : plan.getInputs()) {
> if (!input.needsLock() || input.isUpdateOrDelete()) {
> // We don't want to acquire readlocks during update or delete as
> we'll be acquiring write
> // locks instead.
> continue;
> }
> {noformat}
> whatever sets isUpdateOrDelete() flag doesn't pay attention to whether the
> table is written to or not.
> HIVE-10150 was a similar issue, abstractly
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
