Mailing-List: contact core-dev-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: core-dev@hadoop.apache.org
Message-ID: <334798386.1221699944201.JavaMail.jira@brutus>
Date: Wed, 17 Sep 2008 18:05:44 -0700 (PDT)
From: "Konstantin Shvachko (JIRA)" <jira@apache.org>
To: core-dev@hadoop.apache.org
Subject: [jira] Commented: (HADOOP-4198) DFSClient should do lease recovery
 using data transfer port.
In-Reply-To: <957264628.1221696344195.JavaMail.jira@brutus>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/HADOOP-4198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12632042#action_12632042 ] 

Konstantin Shvachko commented on HADOOP-4198:
---------------------------------------------

You still have the implementation of ClientDatanodeProtocol.recoverBlock in the DataNode class.
I think it makes sense to remove and move the logic (which is just print a log massage) to the other DataNode.recoverBlock() method.
This should make things simpler.

> DFSClient should do lease recovery using data transfer port.
> ------------------------------------------------------------
>
>                 Key: HADOOP-4198
>                 URL: https://issues.apache.org/jira/browse/HADOOP-4198
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: dfs
>    Affects Versions: 0.18.0
>            Reporter: Konstantin Shvachko
>            Assignee: Tsz Wo (Nicholas), SZE
>            Priority: Blocker
>             Fix For: 0.18.1
>
>         Attachments: 4198_20080917.patch
>
>
> HADOOP-3283 introduced {{dfs.datanode.ipc.address}} which defines where the data-node rpc server runs.
> The rpc server on a data-node is used only for lease recovery (HADOOP-3310).
> Lease recovery can be initialized by a name-node or by a client.
> The problem was reported if lease recovery is initialized by a client running on an untrusted host.
> The port that the http server runs on is closed for the outside use and therefore lease recovery fails.
> Production level security model assumes that data-nodes are run on trusted nodes and therefore it is safe to have ports open for inter data-node communication.
> HDFS clients can run on arbitrary nodes and according to the security model can access only the ports that are externally open.
> We propose to use the standard data node port for lease recovery, which means that lease recovery will use {{DataXceiver}} and data transfer protocol rather than {{ClientDatanodeProtocol}}.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.