hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Matt Foley <mfo...@hortonworks.com>
Subject Re: [VOTE] Hadoop release candidate 1.2.0-rc1
Date Tue, 14 May 2013 00:35:00 GMT
Thanks for the reference.

Roy's email clearly says that the thing to be voted on should be source
only. This email is in the context of a discussion about a release
candidate that incorporated jars (from a third-party project) WITHOUT
sources.  In particular, the offending project had apparently captured
certain object files that were depended upon, and included them in the
"source" repository.  This is not what Hadoop builds do.

The document http://www.apache.org/dev/release.html which is stated to be
normative, says,

All releases are in the form of the source materials needed to make changes
to the software being released. In some cases, binary/bytecode packages are
also produced as a convenience to users that might not have the appropriate
tools to build a compiled version of the source. In all such cases, the
binary/bytecode package must have the same version number as the source
release and may only add binary/bytecode files that are the result of
compiling that version of the source code release.


And the document
http://incubator.apache.org/guides/releasemanagement.html#best-practicewhich
is referenced multiple times in Roy's email (although it states
itself to be non-normative), obviously contemplates that binaries may be
released along with the sources, because in the section about the "release
artifacts distribution directory" it says,

Many projects [optionally] use structured layouts... The common theme is
that each type of artifact is grouped into a subdirectory. For example,
binary packages into binaries and source packages into source (say).


So it is very clear that we may continue producing convenience binary
artifacts, as long as they are a result of and of identical provenance as
the sources.
And I hope we all understand that the file hadoop-1.2.0-bin.tar.gz met that
criteria, and was only for convenience and was not the subject of the vote.

However, the file hadoop-1.2.0.tar.gz, which was the subject of the vote,
included both source (full, buildable source), and a corresponding set of
built artifacts produced from that source on a CentOS 6 platform under my
control per (in the normative document)
http://www.apache.org/dev/release.html#owned-controlled-hardware .  Once
again, it was the intent that only the sources were the subject of the
vote, and the binaries themselves are clearly of the kind anticipated by
these policies.

BUT if the fact of packaging them together invalidated the vote, I will
re-create it and run a vote again.

Regardless, I will going forward change the build.xml file to produce a
pure source tarball so that can be the unambiguous subject of the vote.

Roy, if you're listening in, can you please say whether I need to re-do
this vote?

Thanks,
--Matt



On Mon, May 13, 2013 at 4:32 PM, Chris Douglas <cdouglas@apache.org> wrote:

> FWIW: http://s.apache.org/nnN
>
> The rest of the thread (and related discussion that month) is fairly
> unambiguous about what the PMC is allowed to approve. Elsewhere,
> there's clarification that the prohibition is against binaries for
> which we don't also distribute source, so (AFAICT) distributing
> third-party jars is also not kosher. I'll ask for clarification. -C
>
> On Mon, May 13, 2013 at 2:44 PM, Matt Foley <mfoley@hortonworks.com>
> wrote:
> > Hmm.  My understanding was that only sources constituted a "release" and
> > that all release votes were to be understood as votes on a body of source
> > code. However, we've always (at least for the last 2+ years that I've
> been
> > involved in the RM side) distributed binary tarball (and often rpms and
> > debs), ALONG WITH the source tarball, for the convenience of our many
> users
> > who don't care to do builds before using a release.  The binaries and
> > source-containing artifacts are all signed for tamper-resistance, and
> when
> > a Release Manager distributes a set of stuff, the binaries should be
> > understood to come from the same build as the source tarball -- as is
> > indeed the case here.
> >
> > Furthermore, I believe the Hadoop-related projects make use of a Maven
> > server.  I don't believe it's distributing source only :-)
> >
> > I totally agree that the official release is the sources.  But to go from
> > there to a prohibition on distributing objects would, I think, cripple
> the
> > project, and certainly goes against the tradition of common usage in
> > opensource projects, including many Apache projects.
> >
> > Respectfully,
> > --Matt
> >
> >
> >
> > On Mon, May 13, 2013 at 2:01 PM, Chris Douglas <cdouglas@apache.org>
> wrote:
> >
> >> Thanks, Matt. As always, your work on this is hugely appreciated.
> >>
> >> As I understand it, we can't distribute binary-only artifacts. Among
> >> the reasons: the PMC can't verify binaries as project output, the
> >> non-profit charter is about source code, and users need to be able to
> >> modify what we distribute. I can try to track down a reference, but
> >> I'm pretty sure on this one... source-only is OK. Some have argued
> >> it's the only acceptable form. -C
> >>
> >> On Mon, May 13, 2013 at 1:36 PM, Matt Foley <mattf@apache.org> wrote:
> >> > The vote passed and we have accepted Hadoop version 1.2.0 for release:
> >> > +1 binding: 4 (1 slightly late)
> >> > +1 non-binding: 1
> >> > 0 none
> >> > -1 none
> >> >
> >> > Thanks to all who voted.
> >> > I'll finish publishing the release tonight and announce it to
> general@in
> >> > the morning.
> >> >
> >> > Best regards,
> >> > --Matt
> >> > release manager
> >> >
> >> >
> >> > On Mon, May 13, 2013 at 1:32 PM, Matt Foley <mfoley@hortonworks.com>
> >> wrote:
> >> >
> >> >> Hi Chris,
> >> >> Unless I screwed up my build, hadoop-1.2.0.tar.gz includes the built
> >> >> artifacts as well as full buildable source and docs.
> >> >> Hadoop-1.2.0-bin.tar.gz is intended to contain only the built
> artifacts
> >> >> ("binaries") and deliberately excludes the source
> >> >> and docs, for those who wish a smaller tarball of binaries only.  The
> >> >> artifacts in both are from the same build.
> >> >>
> >> >> So I'll take your +1 on the source tarball :-)
> >> >> Thanks,
> >> >> --Matt
> >> >>
> >> >>
> >> >> On Mon, May 13, 2013 at 1:13 PM, Chris Douglas <cdouglas@apache.org
> >> >wrote:
> >> >>
> >> >>> +1 on hadoop-1.2.0.tar.gz (verified checksums, signature, ran some
> unit
> >> >>> tests)
> >> >>>
> >> >>> but hadoop-1.2.0-bin.tar.gz is missing the source code and can't
be
> >> >>> built. -C
> >> >>>
> >> >>> On Mon, May 6, 2013 at 11:11 AM, Matt Foley <mattf@apache.org>
> wrote:
> >> >>> > Hi all,
> >> >>> > I have posted the signed tarballs for Hadoop 1.2.0-rc1 at
> >> >>> > http://people.apache.org/~mattf/hadoop-1.2.0-rc1/
> >> >>> > Release notes are at:
> >> >>> > releasenotes_1.2.0-rc1.html<
> >> >>>
> >>
> http://people.apache.org/~mattf/hadoop-1.2.0-rc1/releasenotes_1.2.0-rc1.html
> >> >>> >
> >> >>> >
> >> >>> > I'm having a little trouble with Nexus (it seems to have
> forgotten I
> >> >>> exist)
> >> >>> > but am working on that and will post to Nexus as soon as possible.
> >> >>> >
> >> >>> > In the meantime, unless there are objections, I'd like to
start
> the
> >> >>> vote.
> >> >>> > Please review this release candidate and vote it for release.
>  Vote
> >> >>> will end
> >> >>> > in seven days as usual, at 11:30am PDT on Monday 13 May.
> >> >>> >
> >> >>> > Best regards,
> >> >>> > --Matt
> >> >>> > (release manager)
> >> >>>
> >> >>
> >> >>
> >>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message