I don't mind trying to put together a patch for what I described above if
there's a general consensus on the strategy we should take (or at least no
big objections).

I think the multiple jars solution could be troublesome, but maybe after
seeing what a patch looks like for a single jar solution it'll seem nicer.


On Mon, Sep 30, 2013 at 2:45 PM, Cheolsoo Park <piaozhe...@gmail.com> wrote:

> Hi Jeremy,
>
> What you're saying makes sense, and patch is welcome. ;-) But complexity
> comes from that there are many classes that are associated with one
> another, and it seems necessary to bring back all of them together in order
> to provide full backward compatibility.
>
> After spending many hours on the weekend, I concluded that adding more
> workarounds (classes, methods, packages, etc) to the current code makes it
> only less maintainable and readable. So I prefer a simpler approach.
>
> For eg, we can just publish two jars - pig.jar w/ old API and pig-new.jar
> w/ new API - maybe not in 0.12 but in 0.13. Since we already have a
> tez-branch, we can use it to manage the new version of classes. Then, users
> can switch to pig-new.jar gradually in 0.13 and 0.14. When we finally merge
> tez-branch into trunk, we can publish a single jar again.
>
> Of course, this is not trivial either because we have to maintain two
> branches. But I feel that managing two branches independently is easier
> than maintaining all sorts of workarounds for backward compatibility in the
> source code. In addition, we will have more flexibility in terms of
> designing new API because we will be completely free from backward
> compatibility. No?
>
> Thanks,
> Cheolsoo
>
> On Mon, Sep 30, 2013 at 11:12 AM, Jeremy Karn <jk...@mortardata.com>
> wrote:
>
> > What about the option of leaving all of the MR specific logic in the
> > original classes but marking those methods as deprecated and telling
> people
> > to switch to using a MR specific object that extends the original class.
> >  So for example:
> >
> > JobStats - Reverted to being as it was before PIG-3419 but with all MR
> > specific logic deprecated.
> > MRJobStats - Would just extend JobStats.
> >
> > If we did this, external software could switch their code from using
> > JobStats to MRJobStats at their own pace and without breaking against any
> > specific version of Pig.  After a few versions the MR specific logic
> could
> > be removed from JobStats and pushed into MRJobStats and it shouldn't
> break
> > anything for people that had made that change.
> >
> > I'm not familiar with all of the changes in PIG-3419 so this might not
> work
> > everywhere.
> >
> >
> > On Mon, Sep 30, 2013 at 1:43 PM, Cheolsoo Park <piaozhe...@gmail.com>
> > wrote:
> >
> > > To be specific, we will need to revert all the following commits in
> > order:
> > >
> > >
> > > commit ad1b87d4ba073680ad0a7fc8c76baeb8b611c982
> > > Author: Cheolsoo Park <cheol...@apache.org>
> > > Date:   Fri Sep 20 22:47:29 2013 +0000
> > >
> > >     PIG-3471: Add a base abstract class for ExecutionEngine (cheolsoo)
> > >
> > >     git-svn-id:
> > >
> > >
> >
> https://svn.apache.org/repos/asf/pig/trunk@152516513f79535-47bb-0310-9956-ffa450edef68
> > >
> > > commit 4305a6f4737d07396ae13fd95d7c1da7933b38a1
> > > Author: Jianyong Dai <da...@apache.org>
> > > Date:   Wed Sep 18 19:09:49 2013 +0000
> > >
> > >     PIG-3457: Provide backward compatibility for PigStatsUtil and
> > JobStats
> > >
> > >     git-svn-id:
> > >
> > >
> >
> https://svn.apache.org/repos/asf/pig/trunk@152453213f79535-47bb-0310-9956-ffa450edef68
> > >
> > > commit e85cf34c92713aa697a1cda7a9c2b3db139350f7
> > > Author: Cheolsoo Park <cheol...@apache.org>
> > > Date:   Wed Sep 18 15:37:58 2013 +0000
> > >
> > >     PIG-3464: Mark ExecType and ExecutionEngine interfaces as evolving
> > > (cheolsoo)
> > >
> > > commit fd8b7cdf9292b305f02386d560c25298ab492a0b
> > > Author: Cheolsoo Park <cheol...@apache.org>
> > > Date:   Fri Aug 30 20:04:29 2013 +0000
> > >
> > >     PIG-3419: Pluggable Execution Engine (achalsoni81 via cheolsoo)
> > >
> > >     git-svn-id:
> > >
> > >
> >
> https://svn.apache.org/repos/asf/pig/trunk@151906213f79535-47bb-0310-9956-ffa450edef68
> > >
> > >
> > >
> > >
> > > On Mon, Sep 30, 2013 at 10:33 AM, Daniel Dai <da...@hortonworks.com>
> > > wrote:
> > >
> > > > Thanks Cheolsoo! My opinion is provide backward compatibility for
> > > PigStats
> > > > is a must, otherwise it could be a havoc. I imagine PigStats is
> widely
> > > used
> > > > by Pig users via PigRunner and PPNL interface. People use PigStats to
> > > > collect MR job details of the Pig job. Though PigStats is marked for
> > > > Evolving, this is mostly for extending PigStats, not limiting
> PigStats
> > as
> > > > PIG-3419 did. Even if we really need to change, we need to very well
> > > > communicate with users over time, Pig 0.12 is not an option.
> > > >
> > > > PIG-3457 is trying to provide a backward compatibility way for
> > PigStats,
> > > > but just like Cheolsoo said, it is far from ideal. I now tend to
> agree
> > > > Rohini's suggestion on PIG-3419, rollback PIG-3419, until we find a
> > > better
> > > > way. Seems PIG-3419 is a little premature. Besides the above
> mentioned
> > > > PigStats issue, I've already found 2 additional issues:
> > > > 1. "explain" shows unoptimized logical plan instead of optimized one
> > > > 2. HangingJobKiller is removed
> > > >
> > > > How does others think?
> > > >
> > > > Thanks,
> > > > Daniel
> > > >
> > > >
> > > >
> > > >
> > > > On Mon, Sep 30, 2013 at 9:51 AM, Cheolsoo Park <piaozhe...@gmail.com
> >
> > > > wrote:
> > > >
> > > > > Hi devs,
> > > > >
> > > > > PIG-3419 <https://issues.apache.org/jira/browse/PIG-3419> broke
> > > backward
> > > > > compatibility for downstream applications such as Oozie, and
> > > > > PIG-3457<https://issues.apache.org/jira/browse/PIG-3457> is
> > > > > trying to fix it.
> > > > >
> > > > > In summary, we need to keep the old MR-specific JobStats and
> PigStats
> > > for
> > > > > backward compatibility sadly because they're currently exposed in
> > > several
> > > > > user-facing API including PigRunner, PigServer, etc. However, this
> > > > defeats
> > > > > the purpose of PIG-3419
> > > > > <https://issues.apache.org/jira/browse/PIG-3419> because
> > > > > we cannot implement non-MR execution engines in the back-end if the
> > > > > front-end API is tied to MR.
> > > > >
> > > > > Daniel and I were having a discussion in the jira, but it seems
> more
> > > > > complicated than I thought. I am wondering whether anyone has a
> good
> > > > > suggestion on how to solve it. This is blocking the 0.12 release.
> > > > >
> > > > > Thanks,
> > > > > Cheolsoo
> > > > >
> > > >
> > > > --
> > > > CONFIDENTIALITY NOTICE
> > > > 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.
> > > >
> > >
> >
> >
> >
> > --
> >
> > Jeremy Karn / Lead Developer
> > MORTAR DATA / 519 277 4391 / www.mortardata.com
> >
>



-- 

Jeremy Karn / Lead Developer
MORTAR DATA / 519 277 4391 / www.mortardata.com

Reply via email to