hadoop-common-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Xiance SI(司宪策) <adam...@gmail.com>
Subject Complains about new o.a.h.mapreduce.* APIs
Date Sat, 31 Oct 2009 05:57:27 GMT
Hi,

Sorry for starting a new thread. I'm facing the same problem as Ian, see the
quoted
mail below. We used 0.15.0 on our lab cluster before. Recently we are
considering
upgrading to 0.20.x, and notice the significant changes in MapReduce APIs.

To be frankly, *I think the new set of API is a step backward*. First, most
of our old
MR codes break, a lot of warnings on deprecated classes like JobConf, but no
obvious
advantage is brought by the new API. Second, the new APIs favor abstract
classes
over interfaces, see the changes in Mapper, Reducer and OutputCollector.
Lots of our
unit test cases are designed based on mock classes that implements
OutputCollector,
new API just break them all without a better solution for testing (MRUnit is
not ready now,
and it is not convenient for complex test cases) . I tried to
subclass Context, but it
was definitely not a nice experience.

Maybe this is just my case, want to hear from other users about the APIs.

Sincerely,
Xiance Si

On Fri, Jun 5, 2009 at 11:21 PM, Ian Soboroff <ian.soboroff@nist.gov> wrote:

>
> I just started playing with 0.20.0.  I see that the mapred package is
> deprecated in favor of the mapreduce package.  Is there any
> migration documentation for the new API (i.e., something more touristy
> than Javadoc)?  All the website docs and Wiki examples are on the old API.
>
> Sorry if this is on the mailing list... I searched a bit but came up
> dry...
>
> In the same vein, it would be nice if release notes could have some
> narrative to go along with the random assortment of JIRA issue numbers.
> Especially when major API migration is in the works.
>
> Ian
>

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