Hi Sqoop(2) developers,
we're making very good progress in Sqoop 2 land. Current code base still can't move any data around, but I'm sure that we're very close to have something usable and testable soon. With this regard, I would like to propose doing some pre/alpha/beta/unstable/preview/pick-your-own-word-for-unstable-code release as soon as possible to provide early adopters first testable bits. My motivation to provide early bits is to get feedback from actual users as soon as possible, so that we can incorporate good ideas before it will become too expensive.

However releasing such early state raises question what version should we release. I would say that most simple idea is to release version "2.0.0", however due to unstable nature, I believe that it's very bad idea. Hadoop seems to be adding "-alpha" to the version names, for example "2.0.1-alpha"[1]. I don't quite like this solution as I would personally expect that there will be 2.0.1 stable (without "alpha" suffix) that won't have any new features but will be stable soon. As this won't be the case for Sqoop, I would like to propose slightly different solution that I've seen in some other projects. I would like to propose doing release 1.99.1. I believe that the version number implies that it's very far from current stable 1.4 version and very very near to 2.0, but it's not 2.0 yet. I would like to see other opinions about using 1.99.1 version for first cut of our Sqoop 2 branch.

I actually did not invent this idea - I've seen in before. For example KDE project [2] is using very similar approach for quite some time by now as you can see on their archive [3]. Versions like 4.8.90 or 4.8.95 are "pre-releases" for 4.9. Another example would be Ubuntu one client [4].

Jarcec

Links:
1: http://mirror.nexcess.net/apache/hadoop/common/hadoop-2.0.1-alpha/
2: http://kde.org/
3: http://download.kde.org/unstable/
4: https://launchpad.net/ubuntuone-client/stable-4-0

Hi Jarcec,

I like your suggestion of 1.99.x. That seems intuitive and clear what it's
supposed to mean to me.

Thanks,
Cheolsoo

On Fri, Oct 12, 2012 at 1:42 PM, Jarek Jarcec Cecho <[EMAIL PROTECTED]>wrote:

> Hi Sqoop(2) developers,
> we're making very good progress in Sqoop 2 land. Current code base still
> can't move any data around, but I'm sure that we're very close to have
> something usable and testable soon. With this regard, I would like to
> propose doing some
> pre/alpha/beta/unstable/preview/pick-your-own-word-for-unstable-code
> release as soon as possible to provide early adopters first testable bits.
> My motivation to provide early bits is to get feedback from actual users as
> soon as possible, so that we can incorporate good ideas before it will
> become too expensive.
>
> However releasing such early state raises question what version should we
> release. I would say that most simple idea is to release version "2.0.0",
> however due to unstable nature, I believe that it's very bad idea. Hadoop
> seems to be adding "-alpha" to the version names, for example
> "2.0.1-alpha"[1]. I don't quite like this solution as I would personally
> expect that there will be 2.0.1 stable (without "alpha" suffix) that won't
> have any new features but will be stable soon. As this won't be the case
> for Sqoop, I would like to propose slightly different solution that I've
> seen in some other projects. I would like to propose doing release 1.99.1.
> I believe that the version number implies that it's very far from current
> stable 1.4 version and very very near to 2.0, but it's not 2.0 yet. I would
> like to see other opinions about using 1.99.1 version for first cut of our
> Sqoop 2 branch.
>
> I actually did not invent this idea - I've seen in before. For example KDE
> project [2] is using very similar approach for quite some time by now as
> you can see on their archive [3]. Versions like 4.8.90 or 4.8.95 are
> "pre-releases" for 4.9. Another example would be Ubuntu one client [4].
>
> Jarcec
>
> Links:
> 1: http://mirror.nexcess.net/apache/hadoop/common/hadoop-2.0.1-alpha/
> 2: http://kde.org/
> 3: http://download.kde.org/unstable/
> 4: https://launchpad.net/ubuntuone-client/stable-4-0
>

Hi Jarcec,

Thanks for the suggestion, and apologies for the late response. I too feel
that having version number is better than label qualifiers such as alpha.

Overall, I like this proposal and think it will be useful for our
community. I have a couple of optional suggestions to go with it:

1. We create a page on our Wiki that details this scheme so that users can
understand what state a release is in, and
2. Have a criteria established for switching over to the 2.0.0 version
sometime.

Of course, we can address these independent of adopting the version scheme,
so it does not have to be blocked on anything.

Regards,
Arvind Prabhakar

On Fri, Oct 19, 2012 at 10:16 AM, Cheolsoo Park <[EMAIL PROTECTED]>wrote:

> Hi Jarcec,
>
> I like your suggestion of 1.99.x. That seems intuitive and clear what it's
> supposed to mean to me.
>
> Thanks,
> Cheolsoo
>
> On Fri, Oct 12, 2012 at 1:42 PM, Jarek Jarcec Cecho <[EMAIL PROTECTED]
> >wrote:
>
> > Hi Sqoop(2) developers,
> > we're making very good progress in Sqoop 2 land. Current code base still
> > can't move any data around, but I'm sure that we're very close to have
> > something usable and testable soon. With this regard, I would like to
> > propose doing some
> > pre/alpha/beta/unstable/preview/pick-your-own-word-for-unstable-code
> > release as soon as possible to provide early adopters first testable
> bits.
> > My motivation to provide early bits is to get feedback from actual users
> as
> > soon as possible, so that we can incorporate good ideas before it will
> > become too expensive.
> >
> > However releasing such early state raises question what version should we
> > release. I would say that most simple idea is to release version "2.0.0",
> > however due to unstable nature, I believe that it's very bad idea. Hadoop
> > seems to be adding "-alpha" to the version names, for example
> > "2.0.1-alpha"[1]. I don't quite like this solution as I would personally
> > expect that there will be 2.0.1 stable (without "alpha" suffix) that
> won't
> > have any new features but will be stable soon. As this won't be the case
> > for Sqoop, I would like to propose slightly different solution that I've
> > seen in some other projects. I would like to propose doing release
> 1.99.1.
> > I believe that the version number implies that it's very far from current
> > stable 1.4 version and very very near to 2.0, but it's not 2.0 yet. I
> would
> > like to see other opinions about using 1.99.1 version for first cut of
> our
> > Sqoop 2 branch.
> >
> > I actually did not invent this idea - I've seen in before. For example
> KDE
> > project [2] is using very similar approach for quite some time by now as
> > you can see on their archive [3]. Versions like 4.8.90 or 4.8.95 are
> > "pre-releases" for 4.9. Another example would be Ubuntu one client [4].
> >
> > Jarcec
> >
> > Links:
> > 1: http://mirror.nexcess.net/apache/hadoop/common/hadoop-2.0.1-alpha/
> > 2: http://kde.org/
> > 3: http://download.kde.org/unstable/
> > 4: https://launchpad.net/ubuntuone-client/stable-4-0
> >
>

Hi Arvind and Cheolsoo,
thank you very much for your feedback, I really appreciate that. I definitely agree that we should document the versions on the wiki and I'll do it myself once it will become relevant.

I would personally incline to let the discussion when to switch from 1.99.x to 2.y open and simply upgrade number to 2.y when developers will feel it's the right time. I would personally like to see following resolved before going to 2.0.0:

* Have pretty good coverage of most important features available in sqoop 1 - import + export tools. Support export to Hive, HBase and Avro. Support incremental imports, query and table based imports. Support update based export.
* Get clear picture how metadata upgrade will be performed
* Get working oozie integration
* Have a very good test coverage in place
* Do performance tests and performance comparison with Sqoop 1
* Have some beta testing done with some adventurous users to make sure that it's both easily usable and working in pseudo production environments

Jarcec

On Tue, Oct 23, 2012 at 09:38:17AM -0700, Arvind Prabhakar wrote:
> Hi Jarcec,
>
> Thanks for the suggestion, and apologies for the late response. I too feel
> that having version number is better than label qualifiers such as alpha.
>
> Overall, I like this proposal and think it will be useful for our
> community. I have a couple of optional suggestions to go with it:
>
> 1. We create a page on our Wiki that details this scheme so that users can
> understand what state a release is in, and
> 2. Have a criteria established for switching over to the 2.0.0 version
> sometime.
>
> Of course, we can address these independent of adopting the version scheme,
> so it does not have to be blocked on anything.
>
> Regards,
> Arvind Prabhakar
>
> On Fri, Oct 19, 2012 at 10:16 AM, Cheolsoo Park <[EMAIL PROTECTED]>wrote:
>
> > Hi Jarcec,
> >
> > I like your suggestion of 1.99.x. That seems intuitive and clear what it's
> > supposed to mean to me.
> >
> > Thanks,
> > Cheolsoo
> >
> > On Fri, Oct 12, 2012 at 1:42 PM, Jarek Jarcec Cecho <[EMAIL PROTECTED]
> > >wrote:
> >
> > > Hi Sqoop(2) developers,
> > > we're making very good progress in Sqoop 2 land. Current code base still
> > > can't move any data around, but I'm sure that we're very close to have
> > > something usable and testable soon. With this regard, I would like to
> > > propose doing some
> > > pre/alpha/beta/unstable/preview/pick-your-own-word-for-unstable-code
> > > release as soon as possible to provide early adopters first testable
> > bits.
> > > My motivation to provide early bits is to get feedback from actual users
> > as
> > > soon as possible, so that we can incorporate good ideas before it will
> > > become too expensive.
> > >
> > > However releasing such early state raises question what version should we
> > > release. I would say that most simple idea is to release version "2.0.0",
> > > however due to unstable nature, I believe that it's very bad idea. Hadoop
> > > seems to be adding "-alpha" to the version names, for example
> > > "2.0.1-alpha"[1]. I don't quite like this solution as I would personally
> > > expect that there will be 2.0.1 stable (without "alpha" suffix) that
> > won't
> > > have any new features but will be stable soon. As this won't be the case
> > > for Sqoop, I would like to propose slightly different solution that I've
> > > seen in some other projects. I would like to propose doing release
> > 1.99.1.
> > > I believe that the version number implies that it's very far from current
> > > stable 1.4 version and very very near to 2.0, but it's not 2.0 yet. I
> > would
> > > like to see other opinions about using 1.99.1 version for first cut of
> > our
> > > Sqoop 2 branch.
> > >
> > > I actually did not invent this idea - I've seen in before. For example
> > KDE
> > > project [2] is using very similar approach for quite some time by now as
> > > you can see on their archive [3]. Versions like 4.8.90 or 4.8.95 are
> > > "pre-releases" for 4.9. Another example would be Ubuntu one client [4].