[
https://issues.apache.org/jira/browse/HADOOP-1085?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12483800
]
Konstantin Shvachko commented on HADOOP-1085:
---------------------------------------------
= You say that dataNodeFirst cannot be supported, but your code first allows
the parameter and second retains old processing of it.
Data-nodes will not find the name-node if somebody calls the public constructor
with dataNodeFirst=true.
This is a potential bug. I think you should remove the parameter.
= Looking at the name-node code. You are passing the actual name-node ports by
modifying the configuration parameter
of the name-node constructor, which makes it an output parameter. I see three
problems with that approach:
a) it is not documented;
b) if somebody uses a defensive copy (as Nigel calls it) of the config
parameter, meaning somebody who does not trust
the name-node to modify its config, there is no way to get the actual ports
directly from the name-node instance.
c) in a real cluster if I configure fs.default.name = 0 and/or dfs.info.port =
0 then data-nodes will never find the name-node.
I'd propose to create a separate constructor for the NameNode that mini cluster
would use.
I propose not to modify the config inside NameNode but rather use
NameNode.getNameNodeAddress() and FSNamesystem.getNameNodeInfoPort()
inside the mini cluster
> Remove 'port rolling' from Mini{DFS|MR}Cluster
> ----------------------------------------------
>
> Key: HADOOP-1085
> URL: https://issues.apache.org/jira/browse/HADOOP-1085
> Project: Hadoop
> Issue Type: Bug
> Components: test
> Affects Versions: 0.12.0
> Reporter: Arun C Murthy
> Assigned To: Arun C Murthy
> Fix For: 0.13.0
>
> Attachments: HADOOP-1085_20070311_1.patch
>
>
> The rolling of ports in these 2 clusters lead to a lot of timing issues and
> failed test cases; as witnessed in our patch process.
> The way around is to let the OS pick the port for the NameNode/JobTracker and
> let the the DataNode/TaskTracker query them for the port to connect to and
> then use that port.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
