Hi,

Thanks for initiating the discussion. I’d be glad if I were allowed to
present my opinion as one of Hive on Tez users.

I can see we have one immediate problem and one fundamental problem.
First, only a few PMC members are voting for 0.10.4 today. Second, it
is tough to nominate a committer or PMC continuously. I don’t have a
short-term idea of how to resolve the first problem. As for the second
problem, I think Tez should ideally be a part of Hive for some
reasons. Why?

I sometimes feel two kinds of difficulties in making a consistent
effort on Tez. The first one is related to the scope of Tez. The
duties of Tez are as narrow as MapReduce or a part of the core of
Spark/Flink. Though I sometimes want to improve Tez[1], I more often
find areas of improvement about Tez on the Hive side[2]. The second
one is Tez's fantastic stability. I usually don’t see many bugs in
Tez. Because of these two reasons, it is really hard for a single
person to contribute continuously to Tez.

I think Hive’s PMC is knowledgeable enough to accept Tez, I guess ASF
allows Tez PMC to transfer Tez to Hive PMC[3], and I believe Hive
users can benefit from the transfer. I am a bit concerned that it
could make it hard to maintain Tez as an independent library for other
projects, such as Apache Pig, though.

This is just one opinion. Anyway, I am confident that Tez is very
competitive, and we’re actually trying to evolve it more, e.g., by
improving Fair Routing or adding some innovative features. Thanks for
the community’s effort!

- [1] I am working on TEZ-3209 to introduce Fair Routing
- [2] I have found some places to improve integration with Tez, e.g.,
HIVE-23831, HIVE-24485
- [3] At least, PMC is likely to be able to salvage an attic project:
https://attic.apache.org/

Regards,
Okumin

On Fri, Sep 6, 2024 at 9:49 PM László Bodor <bodorlaszlo0...@gmail.com> wrote:
>
> Hi Stamatis!
>
> Thanks for starting this discussion.
> This situation is getting worse, and we still haven't had the chance to
> promote new folks to become PMC. Several aspects here:
>
> 1. Tez is in a local optima in a sense that it's a stable execution engine
> under Hive, and most of the improvements are implemented in Hive nowadays.
> Even though there are open Tez tickets, we don't have large projects -
> however, there is always some stuff - to work on, hence the number of
> potential commits decreased in the last years, making it harder to
> contribute.
> 2. Most of the used-to-be-active PMC members are not active anymore, more
> or less because they left companies that contributed to Hive + Tez the
> most, and they left sooner than we had been able to involve new people.
>
> Some thoughts about the future:
> 1. As long Tez is the default execution engine in Hive, it doesn't make
> sense to move it to the Attic.
> 2. Consider using special project-level by-laws to make a release with
> committer votes, not only PMC member votes <-- I'm afraid it's not possible
> due to ASF by-laws.
> 3. New PMC members: let people become committers, then PMC members with
> fewer commits, by considering their activities with more weight (still:
> commits + code reviews are needed), or considering their activity in e.g.
> Hive Runtime development (which is closely related to Tez, but doesn't
> always involve changes in Tez code)
>
> Regards,
> Laszlo Bodor
>
>
> Stamatis Zampetakis <zabe...@gmail.com> ezt írta (időpont: 2024. szept. 5.,
> Cs, 10:55):
>
> > Hey all,
> >
> > The most recent vote for Apache Tez 0.10.4 has been open for almost a
> > month [1] now. It seems that the project struggles to gather binding
> > votes indicating that the majority of the PMC is inactive (or lost
> > interest). If this situation continues the project will have to move
> > to the Attic [2] and this will be a huge setback for dependent
> > projects especially for Hive.
> >
> > I would like to start some discussions to gather ideas on how we can
> > avoid this situation since Tez is the main execution engine that is
> > used by Hive and there are no immediate plans to replace it.
> >
> > Best,
> > Stamatis
> >
> > [1] https://lists.apache.org/thread/yx5lgnxf7k8k7xvf31k37v9hz01j934f
> > [2] https://attic.apache.org/
> >

Reply via email to