[
https://issues.apache.org/jira/browse/HIVE-15808?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Deepak Jaiswal updated HIVE-15808:
----------------------------------
Description: If there is a semijoin branch on the same operator pipeline
which contains a hash join then it is by design on big table which is not
optimal. The operator cycle detection logic may not find a cycle as there is no
cycle at operator level. However, once Tez builds its task there can be a cycle
at task level causing the query to fail. (was: It is found that the current
logic of cycle detection does not find cycles created when there is a semijoin
branch parallel to a hash join on a reducer.
To avoid such cycles, remove the semijoin reduction optimization.)
Summary: Remove semijoin reduction branch if it is on bigtable along
with hash join (was: Remove Semijoin reduction branch on reducers if there is
hash join)
> Remove semijoin reduction branch if it is on bigtable along with hash join
> --------------------------------------------------------------------------
>
> Key: HIVE-15808
> URL: https://issues.apache.org/jira/browse/HIVE-15808
> Project: Hive
> Issue Type: Bug
> Reporter: Deepak Jaiswal
> Assignee: Deepak Jaiswal
> Attachments: HIVE-15808.patch
>
>
> If there is a semijoin branch on the same operator pipeline which contains a
> hash join then it is by design on big table which is not optimal. The
> operator cycle detection logic may not find a cycle as there is no cycle at
> operator level. However, once Tez builds its task there can be a cycle at
> task level causing the query to fail.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
