[
https://issues.apache.org/jira/browse/HDFS-13653?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16502107#comment-16502107
]
Íñigo Goiri commented on HDFS-13653:
------------------------------------
For randomness, I have some code that checks if it follows the chi distribution
but that's totally overdone.
I would just add a test with a couple namespaces and each of them with say 3
namenodes.
Then one of them would have the random flag and not the other.
Then we can do a loop with say 20 client creations and then check if it's
pointing to one NN or the other.
I would try to avoid using MiniDFSCluster.
> Make dfs.client.failover.random.order a per nameservice configuration
> ---------------------------------------------------------------------
>
> Key: HDFS-13653
> URL: https://issues.apache.org/jira/browse/HDFS-13653
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: federation
> Reporter: Ekanth Sethuramalingam
> Assignee: Ekanth Sethuramalingam
> Priority: Major
> Attachments: HDFS-13653.001.patch, HDFS-13653.002.patch,
> HDFS-13653.003.patch
>
>
> Currently the dfs.client.failover.random.order is applied globally. If we
> have a combination of router and non-router nameservice, the random order
> should ideally be enabled only for the router based nameservice. This Jira is
> to make this configuration per-nameservice so that this can be configured
> independently for each nameservice.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
