[
https://issues.apache.org/jira/browse/AMBARI-17818?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15388482#comment-15388482
]
Swapan Shridhar commented on AMBARI-17818:
------------------------------------------
commits :
trunk:
{code}
commit 2516a059a7ca116d218bb6411fccd4ce11ad59ff
Author: Swapan Shridhar <[email protected]>
Date: Thu Jul 21 14:35:38 2016 -0700
AMBARI-17818. Tez & Hive: Memory management should use scaled down Xmx for
sizing buffers. (Gopal V via Swapan Shridhar)
{code}
branch-2.4:
{code}
commit 72d34c96ecf366d529ead19fffd5e52ef80aae42
Author: Swapan Shridhar <[email protected]>
Date: Thu Jul 21 14:38:58 2016 -0700
AMBARI-17818. Tez & Hive: Memory management should use scaled down Xmx for
sizing buffers. (Gopal V via Swapan Shridhar)
{code}
> Tez & Hive: Memory management should use scaled down Xmx for sizing buffers
> ---------------------------------------------------------------------------
>
> Key: AMBARI-17818
> URL: https://issues.apache.org/jira/browse/AMBARI-17818
> Project: Ambari
> Issue Type: Bug
> Components: stacks
> Reporter: Gopal V
> Assignee: Gopal V
> Attachments: AMBARI-17818.1.patch, AMBARI-17818.2.patch
>
>
> The defaults in Ambari assume 80% of the YARN container size is allocate to
> the user heap (Xmx).
> On top of this particular restriction, the allocation blocks are divided into
> Eden/Survivor/Tenured - a single allocation spanning 40% of the heap might
> fail for tez runtime.io.sort.mb.
> Scale down the Tez sort size and map join by 0.8x to match Xmx.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
