[
https://issues.apache.org/jira/browse/HADOOP-1908?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12528836
]
dhruba borthakur commented on HADOOP-1908:
------------------------------------------
I think that a datanode should report an error if the next datanode in the
pipeline reported an error. This allows the client to detect if it failed to
create the desired number of replicas.
The recovery from this error is simple. The client abandons the entire block,
fetches a new blockid and block locations from the namenode and restarts
flushing the current block.
> Restructure data node code so that block sending/receiving is seperated from
> data transfer header handling
> ----------------------------------------------------------------------------------------------------------
>
> Key: HADOOP-1908
> URL: https://issues.apache.org/jira/browse/HADOOP-1908
> Project: Hadoop
> Issue Type: Improvement
> Components: dfs
> Affects Versions: 0.14.1
> Reporter: Hairong Kuang
> Assignee: Hairong Kuang
> Fix For: 0.15.0
>
> Attachments: datanode.patch
>
>
> This jira is intended for code sharing. I'd like to have a BlockSender which
> is resposible for reading a block from a disk and writing it to an output
> stream, and a BlockReceiver which receives a block from an input stream,
> writes it to a disk, and possibly writes it a mirror ouput stream to support
> pipeline writes. Block sender and receiver code is independent of any block
> IO protocols so it could be shared by block transfer, block write, block
> read, and future block replacement introduced by data node rebalancing.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
