hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jonathan Hsieh <...@cloudera.com>
Subject Re: Upcoming merge of snapshots branch into trunk. (HBASE-6055 and HABSE-7290)
Date Fri, 11 Jan 2013 04:05:11 GMT
Lars H is the release manager for 0.94 and it is his call for what he
will allow or disallow into it.  If Lars is cool with enis's 'a'
option, I'm fine with it.

I do feel that having to maintain code across 2-4 versions (trunk,
0.96, 0.95, 0.94) is more significantly more painful than dealing with
1.   I am strongly in favor on not consider backporting this to 0.94
until we have it solid and voted into trunk.

wrt to enis's 'b' option using the 0.95 -- I think the plan for that
name is a 0.96 preview.

Here's what I'll do.  I'll re-open the issue and move it out from
under the HBASE-6055 umbrella.  Let's continue discussion on this
topic there.  There are a few isolated risks that could be introduced
and Matteo and I have mentioned some of them there.

We still have should discuss the options for merges.  The original
plan was to merge the offline branch (hbase-6055) into trunk first and
then later the online (hbase-7290) to trunk.  In our testing, most of
the problems we are encountering now have to do with restore and clone
behavior (the simple online snapshot has been surprisingly-to-me
robust).  As this becomes more robust, I'm leaning more and more
towards doing one offline+online merge of the hbase-7290 branch to
trunk.   My question is after we merge to trunk, the plan would be to
merge offline+online to 0.94 right?

Jon.

On Thu, Jan 10, 2013 at 3:20 PM, Enis Söztutar <enis.soz@gmail.com> wrote:
> Hi,
>> It turns out that both Cloudera and Hortonworks have plans to backport
> this to 0.94 in their respective distributions (I don't think that is a
> secret, apologies if it was).
> It seems true :). From my HWX hat, I can say that we are interested in
> backporting snapshots into 0.94, and my apache hat says that if (at least)
> two companies are interested in this, we should do it in an official apache
> branch. Now, having said that, ideally we should not be putting new stuff
> into 0.94, which is a stable branch. On Hadoop, since they are past 1.0,
> they kind of solved this by adding new features in 1.1, 1.2, etc.
>
> I propose either:
>  a) doing an exception for 0.94, and doing the backport there. We can do
> off by default.
>  b) we can do a 0.95 which would basically be 0.94+snapshots.
>
> a) has the advantage of being the easier to maintain one, but main drawback
> would be to introduce possible destabilization and a major feature in the
> middle of stable releases
> b) has the advantage of being cleaner, but then we have to maintain 0.94,
> 0.95 and 0.96.
>
> Enis
>
>
> On Thu, Jan 10, 2013 at 12:53 PM, Andrew Purtell <apurtell@apache.org>wrote:
>
>> Just throwing it out there... If you're still including patch sets in
>> nightlies then one of us could port in the snapshots backport from CDH to
>> ASF.
>>
>> On Thu, Jan 10, 2013 at 12:49 PM, Jonathan Hsieh <jon@cloudera.com> wrote:
>>
>> > As I mentioned on the jira, I can go either way +/-0 -- currently
>> > there is only rpc-related patches that are different between trunk and
>> > 0.94.  This does however mean more overhead from the folks committing
>> > code and testing related to this feature (me, matteo, jesse, ted?),
>> > which had me leaning more -0
>> >
>>
>>
>> --
>> Best regards,
>>
>>    - Andy
>>
>> Problems worthy of attack prove their worth by hitting back. - Piet Hein
>> (via Tom White)
>>



-- 
// Jonathan Hsieh (shay)
// Software Engineer, Cloudera
// jon@cloudera.com

Mime
View raw message