[ 
https://issues.apache.org/jira/browse/HDFS-11450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15888985#comment-15888985
 ] 

Arpit Agarwal commented on HDFS-11450:
--------------------------------------

Thanks for this improvement [~vagarychen]. The patch looks excellent! Comments 
are all nitpicks:
# DFSNetworkTopology#getInstance need not use {{ReflectionUtils.newInstance}}. 
It can call {{new DFSNetworkTopology}}.
# DFSTopologyNodeImpl#childrenStorageInfo should be final.
# In DFSTopologyNodeImpl#remove, the comment {{// if the parent node has no 
children, remove the parent node...}} probably belongs after the synchronized 
block.
# Coding convention. The NetworkTopology class doesn't follow it but we should 
in new classes. e.g.
## Spaces around operators {{n.setLevel(this.level+1)}} etc.
## Space after closing parenthesis in cast {{(DatanodeDescriptor)n}} etc.
# We avoid acronyms for variable names like {{DatanodeDescriptor dd}}. You 
could call it {{dnDescriptor}} instead.

I am still reviewing the test case.

> HDFS specific network topology classes with storage type info included
> ----------------------------------------------------------------------
>
>                 Key: HDFS-11450
>                 URL: https://issues.apache.org/jira/browse/HDFS-11450
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: namenode
>            Reporter: Chen Liang
>            Assignee: Chen Liang
>         Attachments: HDFS-11450.001.patch, HDFS-11450.002.patch
>
>
> This JIRA adds storage type info into network topology.
> More specifically, this JIRA adds a storage type map by extending 
> {{InnerNodeImpl}} to describe the available storages under the current node's 
> subtree. This map is updated when a node is added/removed from the subtree.
> With this info, when choosing a random node with storage type requirement, 
> the search could then decide to/not to go deeper into a subtree by examining 
> the available storage types first.
> One to-do item still, is that, we might still need to separately handle the 
> cases where a Datanodes restarts, or a disk is hot-swapped, will file another 
> JIRA in that case.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to