[
https://issues.apache.org/jira/browse/AMBARI-11986?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Srimanth Gunturi updated AMBARI-11986:
--------------------------------------
Attachment: AMBARI-11986.patch
Now from the set of available hosts we pick the NodeManager with the least
memory for memory calculations, and add that host in _clusterData_ dictionary
as {{referenceNodeManagerHost}} so that other stack-advisors can use the same
host.
> Stack advisor not picking correct reference host consistently
> -------------------------------------------------------------
>
> Key: AMBARI-11986
> URL: https://issues.apache.org/jira/browse/AMBARI-11986
> Project: Ambari
> Issue Type: Bug
> Components: contrib
> Affects Versions: 2.1.0
> Reporter: Srimanth Gunturi
> Assignee: Srimanth Gunturi
> Priority: Critical
> Fix For: 2.1.0
>
> Attachments: AMBARI-11986.patch
>
>
> If a cluster had 100 hosts where first host had 60GB ram and rest of hosts
> had 3GB ram. The stack-advisor code picks the first host (60GB ram) for some
> calculations (like 'ramPerContainer'), and then picks host with minimum ram
> (3GB ram) for 'yarn.scheduler.minimum-allocation-mb'.
> Due to this, the {{yarn.scheduler.minimum-allocation-mb}} is greater than
> {{yarn.scheduler.maximum-allocation-mb}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
