[
https://issues.apache.org/jira/browse/PHOENIX-2477?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15036046#comment-15036046
]
James Taylor commented on PHOENIX-2477:
---------------------------------------
Thanks so much for looking out for us, [~enis]. It really helps when we catch
these things early.
bq. Does Phoenix have an explicit guarantee about binary compatibility of the
underlying HBase versions?
We can't guarantee it until we're off of private HBase APIs, but as you know,
we're working as fast as possible to do this. In the past, the HBase community
has been awesome at coming up with creative ways to allow us to maintain b/w
compat. It's excruciatingly painful when it's broken.
Can your patch here be made to work with both old and new HBase versions? We've
sometimes had to resort to putting a simple interface in front of something
like this and then had the factory method conditionally choose based on the
HBase version.
> ClassCastException in IndexedWALEditCodec after HBASE-14501 (possible
> dataloss)
> -------------------------------------------------------------------------------
>
> Key: PHOENIX-2477
> URL: https://issues.apache.org/jira/browse/PHOENIX-2477
> Project: Phoenix
> Issue Type: Bug
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Priority: Critical
> Attachments: phoenix-2477_v1.patch, phoenix-2477_v2.patch
>
>
> HBASE-14501 fixed the semantics in using the InputStream.available() and the
> interface between the actual Decoder and BaseDecoder.
> Running Phoenix with IndexedWALEditCodec on top of an HBase version
> containing HBASE-14501 now causes silent data loss since the Decoder throws
> {code}
> java.lang.ClassCastException: org.apache.hadoop.hbase.codec.BaseDecoder$PBIS
> cannot be cast to java.io.DataInput
> {code}
> which gets silently ignored from ProtobufLogReader.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
