[
https://issues.apache.org/jira/browse/HDFS-12993?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16316765#comment-16316765
]
Kihwal Lee commented on HDFS-12993:
-----------------------------------
The requirement sounds fine at a high level, but {{getBlocks()}} can already be
expensive. We've seen multi-millisecond processing times. Even if it is
acquiring the read lock, a long lock lowers the throughput.
bq. As of now, dn't have report for locking load(only readlock) when there are
huge blocks.
{{getBlocks()}} is more expensive when there are many smaller blocks. E.g. more
blocks to go through to gather 10GB worth. Doing path exclusion in such
clusters can have a huge performance impact.
Please make this have zero additional cost for the normal cases.
> Add an option to Balancer for excluding the paths
> -------------------------------------------------
>
> Key: HDFS-12993
> URL: https://issues.apache.org/jira/browse/HDFS-12993
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: balancer & mover
> Reporter: Brahma Reddy Battula
> Assignee: Brahma Reddy Battula
> Attachments: HDFS-12993.patch
>
>
> *Usecase:* Users/Customers want to keep the important/frequently used
> blocks(where blocks are not written with favoured nodes and block pinning is
> not enabled) when balancer is triggered.This can be useful datacentric jobs
> which will run frequently.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
