Ming, in its current form, Tez Shuffle Handler is not setup to handle
backwards compatibility with MapReduce. This could be easily added, the
complication is that MapReduce hard-codes its auxiliary shuffle service to
be named "mapreduce_shuffle". In addition, there are no compelling benefits
for using the Tez Shuffle Handler with mapreduce except not having to have
two services running. For a typical setup, yarn should define both
mapreduce_shuffle and tez_shuffle auxiliary services and run them
concurrently.

jeagles

On Tue, May 23, 2017 at 7:28 PM, Ming Ma <[email protected]> wrote:

> +1. Is Tez shuffle handler backward compatible with MR jobs? For a test
> cluster running Tez and MR jobs, if Tez shuffle handler is backward
> compatible, we can configure NM to only load Tez shuffle handler.
> Otherwise, will need to configure NM to load both Tez and MR shuffle
> handlers.
>
> On Tue, May 23, 2017 at 9:16 AM, Eric Badger <[email protected]
> >
> wrote:
>
> > +1 (non-binding) given the test failure followup jira ([TEZ-3735] Test
> > failures in TestTaskAttempt and TestAMContainerMap - ASF JIRA)
> > Eric
> >
> > |
> > |   |
> > [TEZ-3735] Test failures in TestTaskAttempt and TestAMContainerMap - ASF
> > JIRA
> >    |  |
> >
> >   |
> >
> >
> >
> >
> >     On Monday, May 22, 2017 3:46 PM, Jason Lowe
> > <[email protected]> wrote:
> >
> >
> >  +1, looking forward to this functionality being available in an official
> > release!
> > Jason
> >
> >
> >     On Friday, May 19, 2017 3:16 PM, Jonathan Eagles <[email protected]>
> > wrote:
> >
> >
> >  This vote is to merge the Tez Shuffle Handler feature branch (TEZ-3334)
> > into the master branch. Having done extensive testing and thanks to all
> the
> > feedback given, the Tez Shuffle Handler delivers on its promise to reduce
> > shuffle times for the auto-reduce case and pave the way for more shuffle
> > types in the future.
> >
> > Please view a patch of the proposed merge attached to the TEZ-3334 jira
> > https://issues.apache.org/jira/browse/TEZ-3334
> >
> > As to be the most useful and time efficient, please test and vote early
> if
> > you have any misgivings or questions about the readiness of the branch so
> > as to not to delay the merge process and increase the amount of work on
> my
> > part to continue to maintain this branch with master.
> >
> > This vote will be open until Wednesday around the same time.
> >
> > Here is my +1.
> >
> > Regards,
> > jeagles
> >
> >
> >
> >
> >
> >
>

Reply via email to