[
https://issues.apache.org/jira/browse/HDFS-9318?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kuhu Shukla updated HDFS-9318:
------------------------------
Attachment: HDFS-9318-v1.patch
The simpler way is to make this threshold check have a configurable factor that
multiplies to the loadAverage instead of the fixed 2x.
> considerLoad factor can be improved
> -----------------------------------
>
> Key: HDFS-9318
> URL: https://issues.apache.org/jira/browse/HDFS-9318
> Project: Hadoop HDFS
> Issue Type: Bug
> Affects Versions: 2.6.0
> Reporter: Kuhu Shukla
> Assignee: Kuhu Shukla
> Attachments: HDFS-9318-v1.patch
>
>
> Currently considerLoad avoids choosing nodes that are too active, so it helps
> level the HDFS load across the cluster. Under normal conditions, this is
> desired. However, when a cluster has a large percentage of nearly full nodes,
> this can make it difficult to find good targets because the placement policy
> wants to avoid the full nodes, but considerLoad wants to avoid the busy
> less-full nodes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
