[
https://issues.apache.org/jira/browse/HDFS-9000?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14935852#comment-14935852
]
Haohui Mai commented on HDFS-9000:
----------------------------------
This leads to a much more general discussion -- have you considered making
rebooting the NN / DN significantly faster instead of making them
reconfigurable?
My biggest concern is that whether the approach is a solid direction given the
complexity of NN / DN. There're around ~700 configurations for NN and DN today.
A lot of questions need to be answered given the scale of the configuration.
For example, what are parts of the configurations that are reconfigurable, what
are the parts that require a full restart? What happens if failures happen
during the process of reconfiguration? What happens if a daemon ended up in an
inconsistent state? Is it even possible to detect the daemon is in a
inconsistent state?
It looks to me that it is much more robust to make restarting much faster
(e.g., persisting information like block reports in DN / NN) instead of making
the daemons reconfigurable.
> Allow reconfiguration without restart for parameters where applicable.
> ----------------------------------------------------------------------
>
> Key: HDFS-9000
> URL: https://issues.apache.org/jira/browse/HDFS-9000
> Project: Hadoop HDFS
> Issue Type: Improvement
> Affects Versions: 2.7.0
> Reporter: Xiaobing Zhou
> Assignee: Xiaobing Zhou
>
> Many parameters can be re-configured without requiring a restart.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
