On Mar 13, 2012, at 9:29 AM, Arvind Prabhakar wrote:
> 
> On Tue, Mar 13, 2012 at 7:53 AM, Ralph Goers <[email protected]> 
> wrote:
>> 
>> On Mar 12, 2012, at 10:59 PM, Juhani Connolly wrote:
>>> 
>>> While we have added some cool new features, and fixed some bugs, other new 
>>> ones have appeared and some of the sub-tasks of the original flume-728 task 
>>> have been left by the wayside. We're talking about putting 1.1.0 out but 
>>> there's no target for what 1.1.0 is meant to have? I hope we're not heading 
>>> towards a development style of "release whatever we've done" without a plan.
>> 
>> Software is never perfect. Apache philosophy has always been that it is 
>> better to release often than constantly be waiting for features or specific 
>> bug fixes - unless they are determined to be blockers.  Users really like 
>> getting their hands on the software without having to build it.
>> 
>> While I think roadmaps are great as a discussion point, the proof of what is 
>> needed is what the developers actually write and commit. It doesn't make 
>> sense to wait for something on the roadmap that no one really wants to 
>> implement or that users are OK with living without.
>> 
>> That said, every attempt should be made to release software that does what 
>> it says it does and is reliable. That is why I asked the question about the 
>> version number.  A release with just a number carries certain expectations 
>> that a release with "beta" in it doesn't have.
> 
> I agree with Ralph's point here. We have had a 1.0.0 alpha release and
> what we release right now is going to be more of a beta than anything
> else. Primarily because of all the reasons Juhani pointed out - this
> release will leave room for improvement.
> 
> +1 on officially labeling 1.1.0-incubating version as a "beta".
> 
> We can continue to have the roadmap discussion and work out the plan
> for delivering the in-flight features in 1.2.0 which will should
> target to make a stable release.

+1 on a 1.1.0 beta. As long as we put in some critical build and stability 
fixes for bugs that crept in recently (FLUME-1027 & FLUME-1028), the state of 
Flume NG today is way better than it was at 1.0.0.

I agree with Juhani that we should pick up the discussion about where we are 
going. After reflecting a bit more on it, I'm not convinced that it would be 
effective to try and lock in what should go in the "next release" though, 
because releasing on a short schedule, say once a month, helps keep us honest 
about where we are and helps us eat the elephant one byte at a time (har har). 
Ralph you are right, few people want to build from source and deploy that to 
production. It's a lot of commitment to knowing what's going on with the state 
of the source tree.

Still, I think it is important for Flume for the developers to stay in sync 
with what each other are doing and try to move in the same direction. I think 
we should also collaborate on solving shared pain points of interest (e.g. 
FileChannel) if it's practical to work together on it.

Best,
Mike

Reply via email to