Thanks for highlighting my concern, Matt. I do not profess to great wisdom in the mechanisms of planning for, developing and putting out stable releases of Hadoop.
My point of view is quite simple, and pardon me for stating the obvious: A thriving Hadoop community is more that the developers of core HDFS and YARN - it includes the users of Hadoop, it includes the developers who write applications for Hadoop, it includes projects such as HBase that build on top of Hadoop, it includes commercial vendors who package Hadoop as part of their Big Data solutions, it includes vendors such as ourselves who build proprietary functionality on top of Hadoop, it includes OS vendors who want to bundle Hadoop as part of their Operating Systems. Again - stating the obvious: For such a community to thrive and grow, a limited number of stable releases, along with a limited number of development branches where strong multi-company development takes place is crucial. There are a number of alternative technologies such as riak, mongoDB, memcache, MapR, etc. that are well funded and eager to overtake Hadoop, should Hadoop fragment. Hopefully, the collective wisdom of the Hadoop PMC will play the benevolent dictator role that Linus plays so well for Linux. I will stop here, since nobody ever reads more than four sentences of any email.... Jagane -----Original Message----- From: Matt Foley [mailto:mfo...@hortonworks.com] Sent: Wednesday, May 22, 2013 12:51 PM To: common-dev@hadoop.apache.org Subject: Re: [PROPOSAL] change in bylaws to remove Release Plan vote Hi Jagane, My response to your concerns is that I hope the PMC will have enough wisdom not to pass votes for a confusing number of releases -- if only to avoid the kind of fragmentation you point out could happen. To date, however, this does not seem to have been a major problem in our community. Indeed, lack of regularity in release schedules is more often cited as a problem. (Which this amendment is orthogonal to, so please start a different discussion thread if anyone wants to get into that issue! :-) Thanks, --Matt On Wed, May 22, 2013 at 3:52 AM, Steve Loughran <ste...@hortonworks.com>wrote: > On 21 May 2013 23:47, Jagane Sundar <jag...@sundar.org> wrote: > > > I see one significant benefit to having Release Plan votes: Fewer > releases > > with more members of the community working on any given release. > > In turn, fewer Hadoop releases implies less confusion for end users > > attempting to download and use an Apache Hadoop release. > > > > If there are a dozen different releases of Apache Hadoop available > > for download at the Apache Hadoop website, end users will go to a > > commercial vendor packaged version of Hadoop. That is not good for > > the Apache Hadoop community as a whole. > > > > Jagane > > > > I agree we don't want fragmentation; you don't want to have to choose > between hadoop-2.1, hadoop-2.1.stevel-may and hadoop-2.1.stevel-june. > > With a vote on artifact releases, this can be prevented. I am free to > create my -may and -june artifacts, but the PMC -it is just the PMC > right?- get to say "no steve, you can't ship this from the apache.org" > site, though I am free to make my own (which I have done in the past & > put into my own RPMs. No need for a vote if I do it on my own site, > though I did make sure I named the JARs and RPMs something else so > that maven builds didn't get confused. >
