hadoop-general mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rajiv Chittajallu <raj...@yahoo-inc.com>
Subject Re: 0.23.3 release coming soon
Date Fri, 24 Aug 2012 18:23:44 GMT
----- Original Message -----

> From: Inder.dev Java <java.inder@gmail.com>
> To: general@hadoop.apache.org
> Cc: 
> Sent: Friday, August 24, 2012 8:16 AM
> Subject: Re: 0.23.3 release coming soon
> 
>> We don't have a private git repo.  That was the whole point of this so 
> we
>> can do all of this in public.  What we run and build is purely what is on
>> branch-0.23.
> 
>   True, you can do this by simply sharing the build somewhere and publish
> the link in list.
>   whoever wants they may try.
> 
>   Let's please don't deviate the release numberings by keeping this
> releases somewhere in apache release folder along with Hadoop-2 and 1.
> 
> Already we may have to answer to the people why we skipped 22 version.

What did you do with 0.17?

I don't see an reason for someone to be concerned about skipping 0.22 and 0.21 release. We
moved beyond it. Couple of lines in the release twiki should suffice. I run clusters for reasonably
varied users, this hasn't been an issue. 

When the decision was made to use HA branch as for 2.0 instead of base 0.23 branch, there
was expectation that we would move to stabilize 0.23 by Feb 2012 and move to 2.0 in April.
But unfortunately it took longer than expected. There is lot of 
effort went in to getting 0.23 in to a state where we can start deploying to production. 

> I know many people maintain the source code by cutting some branch. I don't
> think people will ask to make release of their branches to make sure no
> private changes ;-) . They migrate to near latest version to make sure they
> are close with community and will have tests to run on it to make sure no
> impacts.

We don't what to go to the same state as with 0.20.1xx. There are quite an number of folks
who what to use a release of what Yahoo! runs in production. In the last 5 years, with the
exception of some 0.20 release, we published via apache. 

Its up to the PMC to decide whether to publish or not to.


> -thx
> 
> 
> On Fri, Aug 24, 2012 at 4:33 AM, Robert Evans <evans@yahoo-inc.com> wrote:
> 
>>  We don't have a private git repo.  That was the whole point of this so 
> we
>>  can do all of this in public.  What we run and build is purely what is on
>>  branch-0.23.
>> 
>>  > Note that, end of the day, ultimately community has to mark hadoop-2 
> as
>>  >stable. but not 23.3..versions right.
>> 
>>  The community can do what ever they want following the Hadoop Project's
>>  bylaws http://hadoop.apache.org/bylaws.html.  An official release of
>>  hadoop requires a lazy majority vote of PMC members.  Marking a release as
>>  stable is not something that is voted on according to the bylaws.  We all
>>  want hadoop-2 to become stable, but it is up to individual users if it is
>>  stable enough for them.  The fact that hadoop-2.0.0 and hadoop-2.1.0 are
>>  marked as alpha is only to warn people that the release has not really
>>  been tested at scale.  There was a lot of discussion when the alpha was
>>  added to the name if that really was necessary.  The community decided
>>  that it was helpful so that is why we did it.
>> 
>>  0.23.3 is very close to 2.0.0-aplha but without Name Node HA, and with
>>  more bug fixes.  You are correct that even if all of the bugs in 0.23 are
>>  fixed it does not guarantee that all of the bugs in 2 will be fixed
>>  because of that.  But it does guarantee that someone else will not have to
>>  fix a bug in 2 that is also in 0.23.
>> 
>>  --Bobby Evans
>> 
>>  On 8/23/12 4:52 PM, "Inder.dev Java" <java.inder@gmail.com> 
> wrote:
>> 
>>  >> We also wanted to do this out in the open so the community
>>  >could see what was happening, instead of in some private git 
> repository.
>>  >
>>  >You mean, you have some different changes in 23.3,  which is not put 
> into
>>  >hadoop-2 ?
>>  >
>>  >
>>  >alpha cuts already going on and ppl are testing. If you find any bug in
>>  >23.3, you may contribute to hadoop-2.
>>  >That changes will be tested in next alpha cut right. If you test with 
> 23.3
>>  >and say it is stable with out merging some things from hadoop-2, that 
> will
>>  >not really give confidence on hadoop-2 releases right. If you are 
> porting
>>  >everything from hadoop-2 to 23.3, then both releases are same. Need not
>>  >release it separately right? But I am not sure about it.
>>  >
>>  >IMO, having official releases like this may confuse ppl differently.
>>  >
>>  >-thx
>>  >
>>  >
>>  >On Fri, Aug 24, 2012 at 3:04 AM, Robert Evans 
> <evans@yahoo-inc.com>
>>  wrote:
>>  >
>>  >> There was a discussion about this in April
>>  >>
>>  >>
>>  >>
>>  http://mail-archives.apache.org/mod_mbox/hadoop-general/201204.mbox/%3CCB
>>  >>9F
>>  >> 2CBB.37CD3%25evans@yahoo-inc.com%3E
>>  >>
>>  >> in which there was unanimous approval of the idea.  I realize that 
> it is
>>  >> confusing and I feel your pain.  I hate having to explain to our
>>  >>customers
>>  >> why we are going "backwards" from 1.0 to 0.23.
>>  >>
>>  >> The community as a whole is working hard and is serious about
>>  >>stabilizing
>>  >> hadoop-2.  As I said in the original proposal everything that is 
> in
>>  >>0.23.3
>>  >> has been merged first to branch-2.
>>  >>
>>  >> What is more, putting on my Yahoo hat, we have nightly builds and 
> tests
>>  >>of
>>  >> branch-2.  We want to move to branch-2 as soon as possible, but to 
> meet
>>  >> our goals to have YARN out in production by the end of this year 
> we had
>>  >>to
>>  >> cut somewhere.  We also wanted to do this out in the open so the
>>  >>community
>>  >> could see what was happening, instead of in some private git 
> repository.
>>  >>
>>  >> If the community has changed their mind and having an official 
> release
>>  >> labeled 0.23 is too confusing to customers I am happy to not call 
> a
>>  >>vote,
>>  >> but simply renumber the release and move on.
>>  >>
>>  >> --Bobby Evans
>>  >>
>>  >> On 8/23/12 4:09 PM, "Inder.dev Java" 
> <java.inder@gmail.com> wrote:
>>  >>
>>  >> >Having releases like this kind of versions (back to 2x.x..) 
> may confuse
>>  >> >people?
>>  >> >Hadoop-2 and 23.x may not have big difference right.
>>  >> >
>>  >> >already community decided to make the branches as 1, 
> 2....etc.. So
>>  >>again
>>  >> >going back to 2x.x releases may really confuse people. I 
> confused by
>>  >> >seeing
>>  >> >this release mail here.
>>  >> >
>>  >> >Why can't people put real effort on hadoop-2 itself and 
> make it stable
>>  >>as
>>  >> >that is the version community wanted to make it stable from 
> alpha? Why
>>  >> >this
>>  >> >effort split here.
>>  >> >
>>  >> >-thx
>>  >> >
>>  >> >On Fri, Aug 24, 2012 at 2:22 AM, Robert Evans 
> <evans@yahoo-inc.com>
>>  >> wrote:
>>  >> >>
>>  >> >>
>>  >> >> I am planning to do cut a release candidate of 0.23.3 in 
> the next
>>  >>week
>>  >> >>or
>>  >> >> so.  I am waiting for a few more JIRAs to go in 
> (MAPREDUCE-3943 and
>>  >> >> HDFS-3177), and a bit more testing before I call a vote.  
> I know not
>>  >>too
>>  >> >> many people have been putting things into branch-0.23 
> directly, but
>>  >>if
>>  >> >>you
>>  >> >> do want to please check with me first. After I cut the 
> release I will
>>  >> >>open
>>  >> >> the branch up again for more bug fixes for a 23.4 
> release.  I expect
>>  >> >>0.23.4
>>  >> >> to come about a month after 0.23.3. This is to 
> incorporate any bug
>>  >>fixes
>>  >> >> needed as Yahoo! and hopefully others do more testing on 
> 0.23.3 and
>>  >> >>start
>>  >> >> rolling it out to more clusters.
>>  >> >>
>>  >> >> --Bobby Evans
>>  >> >>
>>  >>
>>  >>
>> 
>> 
> 

Mime
View raw message