[
https://issues.apache.org/jira/browse/HDFS-6524?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16950871#comment-16950871
]
Ayush Saxena commented on HDFS-6524:
------------------------------------
bq. Looks like a good improvement, thanks Lisheng Sun.
[~weichiu] This doesn't seems to be an improvement to me, This seems wrong fix
to me, we can't do this change, this failures has no relation with replication
count, it is the count for refetching block locations for Namenode, which is
independent of number of replicas.
I am not sure what I am missing.
[~surendrasingh] can you also give a check once!!!
> Choosing datanode retries times considering with block replica number
> ----------------------------------------------------------------------
>
> Key: HDFS-6524
> URL: https://issues.apache.org/jira/browse/HDFS-6524
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: hdfs-client
> Affects Versions: 3.0.0-alpha1
> Reporter: Liang Xie
> Assignee: Lisheng Sun
> Priority: Minor
> Labels: BB2015-05-TBR
> Attachments: HDFS-6524.001.patch, HDFS-6524.002.patch,
> HDFS-6524.003.patch, HDFS-6524.004.patch, HDFS-6524.005(2).patch,
> HDFS-6524.005.patch, HDFS-6524.006.patch, HDFS-6524.007.patch, HDFS-6524.txt
>
>
> Currently the chooseDataNode() does retry with the setting:
> dfsClientConf.maxBlockAcquireFailures, which by default is 3
> (DFS_CLIENT_MAX_BLOCK_ACQUIRE_FAILURES_DEFAULT = 3), it would be better
> having another option, block replication factor. One cluster with only two
> block replica setting, or using Reed-solomon encoding solution with one
> replica factor. It helps to reduce the long tail latency.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
