I like the idea of Avro as a TLP.
Avro has appeal to groups outside of the Hadoop community even though Hadoop
will likely be our biggest "customer" for some time. Our release cycle and
decisions should be driven by this broader user community. The fact that
Avro is a subproject of Hadoop can imply that Avro is "Hadoop RPC" or that
it "requires" Hadoop or that it's only useful for huge datasets. I think
having Avro as a separate TLP would help disabuse people of those notions.
There is a lot of buzz around Hadoop now and leaving could have a negative
effect on our visibility. However, I think we'll have no trouble generating
our own buzz. We can, for example, develop strong benchmarks that show we
beat other serialization/RPC systems in performance and announce it loudly.
We already have a generous Apache license. If we make Avro an easy drop-in
replacement for other systems, we'll likely see broad adoption. Lots of
happy Avro consumers == buzz.
On Wed, Mar 17, 2010 at 2:09 PM, Doug Cutting <[EMAIL PROTECTED]> wrote:
> I'd like to start a discussion about promoting Avro from a Hadoop
> sub-project to a top-level Apache project (TLP).
> This is not yet a vote. Once we have established general understanding and
> agreement, I'll call a vote.
> I propose we move Avro from hadoop.apache.org/avro to avro.apache.org.
> Avro would then have it's own Project Management Committee (PMC) so that it
> can elect committers and create releases on its own. Currently these
> actions require votes by the Hadoop PMC. However I think Avro now has a
> sufficiently large, diverse and distinct community that it can fend for
> I suggest that initial Avro PMC consist of all active Avro committers at
> the time we make the formal proposal. This is typical for new TLPs.
> (Subsequently PMCs tend to promote committers to the PMC. The Hadoop PMC
> generally promotes committers to the PMC after a year of consistent
> activity, while some projects immediately add new committers to their PMC.
> But we don't need to decide our policy for new PMC membership now, only the
> makeup of the initial PMC.)
> I nominate myself as the initial chair of the Avro PMC, with the proviso
> that we adopt a policy of regular chair replacement. I suggest that Avro
> PMC chairs serve a one or two-year term. A PMC chair has no more power than
> other PMC members, but rather has a few more duties. In particular, the
> chair must submit written quarterly reports to the board describing the
> health of the projects developer community. The chair also maintains
> subversion permissions and committer account creation.
> Do these proposal sound reasonable? Any improvements or questions?
> For some background, please read:
> The steps I imagine are roughly:
> - discussion by Avro developers (what I'm starting here today)
> - vote by Avro developers
> - discussion by Hadoop PMC
> - vote by Hadoop PMC
> - draft resolution sent to board
> - board votes on resolution to form TLP
> Formally, the board alone has the power to create a TLP: all prior steps
> are merely an ordered means to make the case to the board that all involved
> parties support such an action.