[
https://issues.apache.org/jira/browse/HDFS-5778?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13909258#comment-13909258
]
Fengdong Yu commented on HDFS-5778:
-----------------------------------
[~szetszwo] I have some confusion after read the document.
{code}
+ <subsection name="Upgrade without Downtime" id="UpgradeWithoutDowntime">
+ <p>
+ In a HA cluster, there are two or more NameNodes (NNs), many DataNodes
(DNs),
+ a few JournalNodes (JNs) and a few ZooKeeperNodes (ZKNs).
+ Note that JNs and ZKNs are not considered in the upgrade procedure
described here.
+ It may incur cluster downtime for upgrading these nodes.
+ </p>
{code}
How to rolling upgrade JNs? if we've upgraded NNs and DNs, then we cannot leave
JNs as old. can you add some descriptions here?
I don't think ZKNs need to upgrade after NNs and DNs are upgraded.
{code}
+ <li>Upgrade DNs<ol>
+ <li>Choose a small subset of datanodes (e.g. all datanodes under a
particular rack).</li>
+ <li>Use <a href="#dfsadminCommands">dfsadmin</a> to shutdown all the
chosen datanodes.
+ Upgrade and restart them.</li>
+ <li>Repeat the above steps until all datanodes in the cluster are
upgraded.</li>
+ </ol></li>
{code}
why choose a small subset DNs to upgrade instead of "dfsadmin -rollingUpgrade
start" for all DNs? maybe there are some special reasons omitted?
> Document new commands and parameters for improved rolling upgrades
> ------------------------------------------------------------------
>
> Key: HDFS-5778
> URL: https://issues.apache.org/jira/browse/HDFS-5778
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: documentation
> Affects Versions: HDFS-5535 (Rolling upgrades)
> Reporter: Akira AJISAKA
> Assignee: Tsz Wo (Nicholas), SZE
> Attachments: h5778_20140220.patch, h5778_20140221.patch
>
>
> "hdfs dfsadmin -rollingUpgrade" command was newly added in HDFS-5752, and
> some other commands and parameters will be added in the future. This issue
> exists to flag undocumented commands and parameters when HDFS-5535 branch is
> merging to trunk.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
