hadoop-yarn-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sandy Ryza <sandy.r...@cloudera.com>
Subject Re: Hadoop Code of Incompatible Changes
Date Tue, 29 Jul 2014 16:39:45 GMT
Hi Zhijie,

The Hadoop compatibility guide mentions this as "semantic compatibility":

My interpretation of the section is that we can't change the behavior of
public APIs unless we're fixing buggy behavior.  If the change could break
an existing application that's behaving reasonably with respect to the old
API, it's an incompatible change.


On Tue, Jul 29, 2014 at 9:26 AM, Zhijie Shen <zshen@hortonworks.com> wrote:

> Hi folks,
> Recently we have a conversation on YARN-2209 about the incompatible changes
> over releases. For those API changes that will break binary compatibility,
> source compatibility towards the existing API users, we've already had a
> rather clear picture about what we should do. However, YARN-2209 has
> introduced another case which I'm not quite sure about, which is kind of
> *logic
> incompatibility*.
> In detail, an ApplicationMasterProtocol API is going to throw an exception
> which is not expected before. The exception is a sub-class of
> YarnException, such that it doesn't need any method signature change, and
> won't break any binary/source compatibility. However, the exception is not
> expected before, but needs to be treated specially at the AM side. Not
> being aware of the newly introduced exception, the existing YARN
> applications' AM may not handle it exception properly, and is at the risk
> of being broken on a new YARN cluster after this change.
> An additional thought around this problem is that the change of what
> exception is to throw under what situation may be considered as a *soft
> *method
> signature change, because we're supposed to write this javadoc to tell the
> users (though we didn't it well in Hadoop), and users refer to it to guide
> how to handle the exception.
> In a more generalized form, let's assume we have a method, which behaves as
> A, in release 1.0. However, in release 2.0, the method signature has kept
> the same, while its behavior is altered from A to B. A and B are different
> behaviors. In this case, do we consider it as an incompatible change?
> I think it's somewhat a common issue, such that I raise it on the mailing
> list. Please share your ideas.
> Thanks,
> Zhijie
> --
> Zhijie Shen
> Hortonworks Inc.
> http://hortonworks.com/
> --
> NOTICE: This message is intended for the use of the individual or entity to
> which it is addressed and may contain information that is confidential,
> privileged and exempt from disclosure under applicable law. If the reader
> of this message is not the intended recipient, you are hereby notified that
> any printing, copying, dissemination, distribution, disclosure or
> forwarding of this communication is strictly prohibited. If you have
> received this communication in error, please contact the sender immediately
> and delete it from your system. Thank You.

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message