[
https://issues.apache.org/jira/browse/HDFS-14220?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16809959#comment-16809959
]
Ayush Saxena commented on HDFS-14220:
-------------------------------------
Thanx [~Amithsha] for putting this up.
{quote}If a user needs max of 1 replica per rack then
BlockPlacementPolicyDefault.java modification is needed instead we can add a
property to specify the placement policy and replica value per rack.
{quote}
Actually the rack fault tolerance assumes the chances of failure of a rack are
quite rare, so two racks failing together isn't that much a probable happening
and since the data is written in form of a pipeline
L-rack-->R-rack-n1--R-rack-2, if we put the last one in a third rack, the write
shall also be slow as inter rack transfers shall take more time as compared to
within rack.
Do you have a use case where this can be required?
> Enable Replica Placement Value Per Rack
> ---------------------------------------
>
> Key: HDFS-14220
> URL: https://issues.apache.org/jira/browse/HDFS-14220
> Project: Hadoop HDFS
> Issue Type: New Feature
> Components: datanode
> Affects Versions: 2.7.1
> Reporter: Amithsha
> Priority: Trivial
>
> By default, Replica placement per rack will be taken care by
> BlockPlacementPolicyDefault.java .
> With 2 if conditions
> # numOfRacks <1
> # numOfRacks > 1
> and the placement will happen as 1 on localrack, 2 on remote rack.
> If a user needs max of 1 replica per rack then
> BlockPlacementPolicyDefault.java modification is needed instead we can add a
> property to specify the placement policy and replica value per rack.
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
