hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Siddharth Seth (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-15163) LLAP Finishable state updates may not go to daemons restarted on the same node
Date Tue, 08 Nov 2016 23:49:58 GMT

    [ https://issues.apache.org/jira/browse/HIVE-15163?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15649230#comment-15649230
] 

Siddharth Seth commented on HIVE-15163:
---------------------------------------

Using dynamic ports will avoid this to a large extent. The actual fix is likely to required
a change in Tez to propagate additional node information from the scheduler to the task communicator,
and the task communicator to watch for node failure updates.
https://issues.apache.org/jira/browse/HIVE-15164 changes the default port.

> LLAP Finishable state updates may not go to daemons restarted on the same node
> ------------------------------------------------------------------------------
>
>                 Key: HIVE-15163
>                 URL: https://issues.apache.org/jira/browse/HIVE-15163
>             Project: Hive
>          Issue Type: Bug
>            Reporter: Siddharth Seth
>
> Finishable state updates only go out to nodes when the source state changes. When a node
goes down, and comes back up on the same port - updates which had been sent to the old node
will not be sent to the new node. This could lead to a fragment being in a permanent preemptable
state, and lead tp starvation on busy clusters.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message