[
https://issues.apache.org/jira/browse/HADOOP-6120?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12733331#action_12733331
]
Doug Cutting commented on HADOOP-6120:
--------------------------------------
If, e.g., a class both implements Writable and also either implements
AvroReflectSerializeable or whose package is listed in the config, then which
serialization is used is determined by the order of the serializers in the
config, which could change. Similarly for the Serializeable interface and
JavaSerialization. So its not safe to assume that the Class->Serialization map
is fixed, and we should really be storing at least the serialization's name in
container metadata, and probably also a version (e.g. serialVersionUID for
JavaSerialization). And once we have a mechanism to support that, we can also
store other metadata, like the schema, so that we can read older versions and
generic data.
+1 for removing JavaSerialization from the defaults.
+1 for a separate Jira on serialization metadata. But it would be best to have
such metadata by the 0.21 freeze, in two weeks.
> Add support for Avro types in hadoop
> ------------------------------------
>
> Key: HADOOP-6120
> URL: https://issues.apache.org/jira/browse/HADOOP-6120
> Project: Hadoop Common
> Issue Type: New Feature
> Reporter: Sharad Agarwal
> Assignee: Sharad Agarwal
> Attachments: 6120_v1.patch, 6120_v2.patch, 6120_v3.patch,
> 6120_v4.patch, 6120_v5.patch, HADOOP-6120.patch
>
>
> Support to serialize and deserialize Avro types in Hadoop.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
