hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Todd Lipcon (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-9151) Include RPC error info in RpcResponseHeader instead of sending it separately
Date Thu, 20 Dec 2012 19:47:14 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-9151?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13537307#comment-13537307

Todd Lipcon commented on HADOOP-9151:

bq. I agree that if possible we should avoid breaking the compatibility. But when a cleanup
can make the implementation lot more clear, I prefer making that change. That is the reason
why we use alpha and any one who uses that release should be ready to adopt the changes.

The implementation is slightly more clear, I agree. We should have done it this way to begin
with. But IMO it's really not worth it.

There are now downstream projects building against Hadoop-2 releases such as HBase. If we
break compat between 2.0.2 and a later release, then HBase users will have the messy "make
sure you replace your HDFS client jar inside HBase's lib directory with the exact right version
or else get weird error messages" nonsense come back again.

I'd be +1 on this for 3.0 but not for branch-2, unless you can figure out some way of maintaining
a compatibility path. For example, we could do something like this in 2.0:
- add an optional flag to the request protobuf indicating that the new style exception response
is supported
- mark the embedded exception 
- new server checks the flag: if the flag is there, it sends the new-style (in-PB) exception
response. If it is not there, it uses the old style response for compatibility
- when the new client gets an ERROR response, but exceptionClassName is not found in the response
PB, then it knows it is an old style server, and reads the writables following the response
(like the old protocol)

This would be compatible as follows:
- old client -> old server: obviously compatible
- old client -> new server: doesn't send flag, so new server responds with old response
- new client -> old server: sends flag, but the old server ignores it (since it's part
of a PB, this is automatic). server responds with old style response.
- new client -> new server: sends flag, so new server responds with new style response.

Obviously, all this makes the code more complex. So, I think it would be better to just leave
branch-2 alone, and make the improvement in branch-3/trunk.
> Include RPC error info in RpcResponseHeader instead of sending it separately
> ----------------------------------------------------------------------------
>                 Key: HADOOP-9151
>                 URL: https://issues.apache.org/jira/browse/HADOOP-9151
>             Project: Hadoop Common
>          Issue Type: Sub-task
>            Reporter: Sanjay Radia
>            Assignee: Sanjay Radia
>         Attachments: HADOOP-9151.patch

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

View raw message