Hey Olaf,

Sorry I jumped into some conclusions so quick without actually getting my
hands dirty, which is not professional. Probably those trigger points I
mentioned are not strong enough in technical perspective. But my rationale
in the back is very simple:

Since all the current PMC members do not have much time in development, we
probably do not fully discovered the potential of Apache Bigtop. If
there're contributors who's willing to contribute, we should give them the
chance to earn the merits through out time and maybe, take on the role to
push Bigtop to the next level.

I really value your efforts to try out Hadoop 3. And your points are sound.
So to mitigate the gap here, how about  we have a quick con call to talk to
each other directly and resolve all the gaps once for all?


Hey Ganesh,
Would you be able to setup a con call to discuss the topics that you and
the Arm folks planning to contribute to Bigtop? What's on top of my head:
1. tar.gz packaging, 2. Hadoop 3, 3. Bigtop blueprints.

And, to make the discussion productive, I really like to know the answer of
why those users are stick to tar.gz for deployment. I don't think "because
it's what there're currently using" is a convincing answer.

- Evans


Olaf Flebbe <[email protected]> 於 2019年7月30日 週二 上午4:22寫道:

> Hi,
>
> > 1. People are using docker for packaging and deployment instead of
> rpm/deb.
>
> You can package everything as a container, but it may or may not make
> sense.
>
> > 2. Hadoop 3's messy layout don't fit into rpm/deb conventions. Maybe it's
> > because they think it's the right design for the container era.
>
> There is nothing prepared to run hadoop in a container. BTW it would be
> messy since yarn is supporting compute loads to run in a container. That
> would mean a nested container runtime. Feasible but unnecessary mind
> twisting.
>
> Hadoop3 does fit in rpm/deb, one only would have to change the packaging
> layout to a monolith approach. Can be done, not nice, but feasible.
>
> I am exhausted in explaining the points over and over again. Go and try,
> I tried it. For a secure hadoop service you need a known IP address, and
> a FQDN hostname resolution. You will find ton's of docker bug reports
> like https://github.com/moby/moby/issues/14282 or
> https://github.com/moby/moby/issues/29100 only to pick up two. I used
> lxc to demonstrate setting up a secure hadoop environment at Apache
> Conference in Budapest for a reason.
>
> I am out of this project now. Do whatever you like.
>
> Olaf
>

Reply via email to