[
https://issues.apache.org/jira/browse/HADOOP-12924?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15205856#comment-15205856
]
Zhe Zhang commented on HADOOP-12924:
------------------------------------
Thanks for the thoughts Kai. Simplifying logic is a valid concern. However, if
we have the legacy HDFS-RAID coder in the Hadoop codebase, there's always a
possibility for some user to use it to encode files. Then by looking at file
metadata, there's no way to determine whether the encoding was done with legacy
or new Java coder.
I think we should either add legacy coder as a policy, or take it out from the
codebase and make it an external took only for migrating legacy HDFS-RAID data
into HDFS-EC.
Actually, how could our current legacy coder be used to migrate legacy cluster
data? IIUC HDFS-RAID was developed in Facebook's private branch based on
upstream version 0.21.0. Is it even possible to run Hadoop 3.0 software on
blocks created by 0.21.0? I think an easier way to migrate is to use the
original HDFS-RAID code to read data out and write into HDFS-EC.
> Add default coder key for creating raw coders
> ---------------------------------------------
>
> Key: HADOOP-12924
> URL: https://issues.apache.org/jira/browse/HADOOP-12924
> Project: Hadoop Common
> Issue Type: Sub-task
> Reporter: Rui Li
> Assignee: Rui Li
> Priority: Minor
> Attachments: HADOOP-12924.1.patch
>
>
> As suggested
> [here|https://issues.apache.org/jira/browse/HADOOP-12826?focusedCommentId=15194402&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15194402].
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
