[
https://issues.apache.org/jira/browse/HADOOP-1549?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12592872#action_12592872
]
Chris Douglas commented on HADOOP-1549:
---------------------------------------
There doesn't appear to be an easy way to "update" the Writables in this patch.
If you wanted to create new Writables and deprecate these- VectorWritable
instead of ArrayWritable, for example- then that might be a way to effect this,
but the UTF8 and Text binaries are incompatible and their differences- at first
glance- are undetectable (i.e. unsigned shorts and vints cannot be
distinguished). Similarly, the change to WritableName is incompatible with
SequenceFile::Reader, and- again- SequenceFiles written prior to this patch
must be distinguished by a version bump and appropriate handling code. With a
version bump, the changes to RPC, FileSplit, JobProfile, JobStatus, Task,
TaskStatus, and TaskTrackerStatus are probably fine, since they're meant to be
transient.
The tests are clearly OK, as replacements for ArrayWritable and ObjectWritable
would be, but the transient stuff will require some extra testing/attention.
This seems like a fairly risky incompatible change. It might be a good idea to
spread some of this across several patches/JIRAs.
> All remaining UTF8 data structures in HDFS code should be removed
> -----------------------------------------------------------------
>
> Key: HADOOP-1549
> URL: https://issues.apache.org/jira/browse/HADOOP-1549
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Reporter: dhruba borthakur
> Assignee: Edward J. Yoon
> Fix For: 0.18.0
>
> Attachments: 1549.patch, 1549_v02.patch
>
>
> The UTF8 data structure is deprecated. HADOOP-1283 addressed part of the
> problem. All remaining UTF8 data structurs in HDFS should be removed.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
