[
https://issues.apache.org/jira/browse/HDFS-14061?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16734506#comment-16734506
]
Ayush Saxena commented on HDFS-14061:
-------------------------------------
Thanks [~knanasi] for the patch.
{quote}A warning is shown if the cluster topology verify fails for all enabled
policies, when enabling a policy.
{quote}
I guess if we are enabling RS(6,3) warning should be shown wrt to the policy
which we are enabling.We should not be bothered about the others not in that
transaction.
{quote}Policy setting fails if the cluster topology verification fails for the
policy
{quote}
IMO This will be quite harsh and guess might be considered as incompatible
too.Moreover for EC File say RS(6,3) I guess It is not mandatory that it should
have strictly 9 DN's for successful write. In cases it can be successful with
lesser numbers too.
I think a warning if the policy that we are enabling is not supported should be
well enough.
> Check if the cluster topology supports the EC policy before setting, enabling
> or adding it
> ------------------------------------------------------------------------------------------
>
> Key: HDFS-14061
> URL: https://issues.apache.org/jira/browse/HDFS-14061
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: erasure-coding, hdfs
> Affects Versions: 3.1.1
> Reporter: Kitti Nanasi
> Assignee: Kitti Nanasi
> Priority: Major
> Attachments: HDFS-14061.001.patch
>
>
> HDFS-12946 introduces a command for verifying if there are enough racks and
> datanodes for the enabled erasure coding policies.
> This verification could be executed for the erasure coding policy before
> enabling, setting or adding it and a warning message could be written if the
> verify fails, or the policy setting could be even failed in this case.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
